bump dependencies

This commit is contained in:
raf 2025-02-08 16:32:45 +03:00
commit b7319e6bfc
No known key found for this signature in database
GPG key ID: EED98D11B85A2819
466 changed files with 17912 additions and 12742 deletions

14
vendor/github.com/fsnotify/fsnotify/.cirrus.yml generated vendored Normal file
View file

@ -0,0 +1,14 @@
freebsd_task:
name: 'FreeBSD'
freebsd_instance:
image_family: freebsd-14-1
install_script:
- pkg update -f
- pkg install -y go
test_script:
# run tests as user "cirrus" instead of root
- pw useradd cirrus -m
- chown -R cirrus:cirrus .
- FSNOTIFY_BUFFER=4096 sudo --preserve-env=FSNOTIFY_BUFFER -u cirrus go test -parallel 1 -race ./...
- sudo --preserve-env=FSNOTIFY_BUFFER -u cirrus go test -parallel 1 -race ./...
- FSNOTIFY_DEBUG=1 sudo --preserve-env=FSNOTIFY_BUFFER -u cirrus go test -parallel 1 -race -v ./...

View file

@ -1,12 +0,0 @@
root = true
[*.go]
indent_style = tab
indent_size = 4
insert_final_newline = true
[*.{yml,yaml}]
indent_style = space
indent_size = 2
insert_final_newline = true
trim_trailing_whitespace = true

View file

@ -1 +0,0 @@
go.sum linguist-generated

View file

@ -4,3 +4,7 @@
# Output of go build ./cmd/fsnotify
/fsnotify
/fsnotify.exe
/test/kqueue
/test/a.out

View file

@ -1,16 +1,115 @@
# Changelog
All notable changes to this project will be documented in this file.
1.8.0 2023-10-31
----------------
The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
### Additions
## [Unreleased]
- all: add `FSNOTIFY_DEBUG` to print debug logs to stderr ([#619])
Nothing yet.
### Changes and fixes
## [1.6.0] - 2022-10-13
- windows: fix behaviour of `WatchList()` to be consistent with other platforms ([#610])
- kqueue: ignore events with Ident=0 ([#590])
- kqueue: set O_CLOEXEC to prevent passing file descriptors to children ([#617])
- kqueue: emit events as "/path/dir/file" instead of "path/link/file" when watching a symlink ([#625])
- inotify: don't send event for IN_DELETE_SELF when also watching the parent ([#620])
- inotify: fix panic when calling Remove() in a goroutine ([#650])
- fen: allow watching subdirectories of watched directories ([#621])
[#590]: https://github.com/fsnotify/fsnotify/pull/590
[#610]: https://github.com/fsnotify/fsnotify/pull/610
[#617]: https://github.com/fsnotify/fsnotify/pull/617
[#619]: https://github.com/fsnotify/fsnotify/pull/619
[#620]: https://github.com/fsnotify/fsnotify/pull/620
[#621]: https://github.com/fsnotify/fsnotify/pull/621
[#625]: https://github.com/fsnotify/fsnotify/pull/625
[#650]: https://github.com/fsnotify/fsnotify/pull/650
1.7.0 - 2023-10-22
------------------
This version of fsnotify needs Go 1.17.
### Additions
- illumos: add FEN backend to support illumos and Solaris. ([#371])
- all: add `NewBufferedWatcher()` to use a buffered channel, which can be useful
in cases where you can't control the kernel buffer and receive a large number
of events in bursts. ([#550], [#572])
- all: add `AddWith()`, which is identical to `Add()` but allows passing
options. ([#521])
- windows: allow setting the ReadDirectoryChangesW() buffer size with
`fsnotify.WithBufferSize()`; the default of 64K is the highest value that
works on all platforms and is enough for most purposes, but in some cases a
highest buffer is needed. ([#521])
### Changes and fixes
- inotify: remove watcher if a watched path is renamed ([#518])
After a rename the reported name wasn't updated, or even an empty string.
Inotify doesn't provide any good facilities to update it, so just remove the
watcher. This is already how it worked on kqueue and FEN.
On Windows this does work, and remains working.
- windows: don't listen for file attribute changes ([#520])
File attribute changes are sent as `FILE_ACTION_MODIFIED` by the Windows API,
with no way to see if they're a file write or attribute change, so would show
up as a fsnotify.Write event. This is never useful, and could result in many
spurious Write events.
- windows: return `ErrEventOverflow` if the buffer is full ([#525])
Before it would merely return "short read", making it hard to detect this
error.
- kqueue: make sure events for all files are delivered properly when removing a
watched directory ([#526])
Previously they would get sent with `""` (empty string) or `"."` as the path
name.
- kqueue: don't emit spurious Create events for symbolic links ([#524])
The link would get resolved but kqueue would "forget" it already saw the link
itself, resulting on a Create for every Write event for the directory.
- all: return `ErrClosed` on `Add()` when the watcher is closed ([#516])
- other: add `Watcher.Errors` and `Watcher.Events` to the no-op `Watcher` in
`backend_other.go`, making it easier to use on unsupported platforms such as
WASM, AIX, etc. ([#528])
- other: use the `backend_other.go` no-op if the `appengine` build tag is set;
Google AppEngine forbids usage of the unsafe package so the inotify backend
won't compile there.
[#371]: https://github.com/fsnotify/fsnotify/pull/371
[#516]: https://github.com/fsnotify/fsnotify/pull/516
[#518]: https://github.com/fsnotify/fsnotify/pull/518
[#520]: https://github.com/fsnotify/fsnotify/pull/520
[#521]: https://github.com/fsnotify/fsnotify/pull/521
[#524]: https://github.com/fsnotify/fsnotify/pull/524
[#525]: https://github.com/fsnotify/fsnotify/pull/525
[#526]: https://github.com/fsnotify/fsnotify/pull/526
[#528]: https://github.com/fsnotify/fsnotify/pull/528
[#537]: https://github.com/fsnotify/fsnotify/pull/537
[#550]: https://github.com/fsnotify/fsnotify/pull/550
[#572]: https://github.com/fsnotify/fsnotify/pull/572
1.6.0 - 2022-10-13
------------------
This version of fsnotify needs Go 1.16 (this was already the case since 1.5.1,
but not documented). It also increases the minimum Linux version to 2.6.32.

View file

@ -1,7 +1,7 @@
Thank you for your interest in contributing to fsnotify! We try to review and
merge PRs in a reasonable timeframe, but please be aware that:
- To avoid "wasted" work, please discus changes on the issue tracker first. You
- To avoid "wasted" work, please discuss changes on the issue tracker first. You
can just send PRs, but they may end up being rejected for one reason or the
other.
@ -20,6 +20,124 @@ platforms. Testing different platforms locally can be done with something like
Use the `-short` flag to make the "stress test" run faster.
Writing new tests
-----------------
Scripts in the testdata directory allow creating test cases in a "shell-like"
syntax. The basic format is:
script
Output:
desired output
For example:
# Create a new empty file with some data.
watch /
echo data >/file
Output:
create /file
write /file
Just create a new file to add a new test; select which tests to run with
`-run TestScript/[path]`.
script
------
The script is a "shell-like" script:
cmd arg arg
Comments are supported with `#`:
# Comment
cmd arg arg # Comment
All operations are done in a temp directory; a path like "/foo" is rewritten to
"/tmp/TestFoo/foo".
Arguments can be quoted with `"` or `'`; there are no escapes and they're
functionally identical right now, but this may change in the future, so best to
assume shell-like rules.
touch "/file with spaces"
End-of-line escapes with `\` are not supported.
### Supported commands
watch path [ops] # Watch the path, reporting events for it. Nothing is
# watched by default. Optionally a list of ops can be
# given, as with AddWith(path, WithOps(...)).
unwatch path # Stop watching the path.
watchlist n # Assert watchlist length.
stop # Stop running the script; for debugging.
debug [yes/no] # Enable/disable FSNOTIFY_DEBUG (tests are run in
parallel by default, so -parallel=1 is probably a good
idea).
touch path
mkdir [-p] dir
ln -s target link # Only ln -s supported.
mkfifo path
mknod dev path
mv src dst
rm [-r] path
chmod mode path # Octal only
sleep time-in-ms
cat path # Read path (does nothing with the data; just reads it).
echo str >>path # Append "str" to "path".
echo str >path # Truncate "path" and write "str".
require reason # Skip the test if "reason" is true; "skip" and
skip reason # "require" behave identical; it supports both for
# readability. Possible reasons are:
#
# always Always skip this test.
# symlink Symlinks are supported (requires admin
# permissions on Windows).
# mkfifo Platform doesn't support FIFO named sockets.
# mknod Platform doesn't support device nodes.
output
------
After `Output:` the desired output is given; this is indented by convention, but
that's not required.
The format of that is:
# Comment
event path # Comment
system:
event path
system2:
event path
Every event is one line, and any whitespace between the event and path are
ignored. The path can optionally be surrounded in ". Anything after a "#" is
ignored.
Platform-specific tests can be added after GOOS; for example:
watch /
touch /file
Output:
# Tested if nothing else matches
create /file
# Windows-specific test.
windows:
write /file
You can specify multiple platforms with a comma (e.g. "windows, linux:").
"kqueue" is a shortcut for all kqueue systems (BSD, macOS).
[goon]: https://github.com/arp242/goon
[Vagrant]: https://www.vagrantup.com/

View file

@ -1,29 +1,31 @@
fsnotify is a Go library to provide cross-platform filesystem notifications on
Windows, Linux, macOS, and BSD systems.
Windows, Linux, macOS, BSD, and illumos.
Go 1.16 or newer is required; the full documentation is at
Go 1.17 or newer is required; the full documentation is at
https://pkg.go.dev/github.com/fsnotify/fsnotify
**It's best to read the documentation at pkg.go.dev, as it's pinned to the last
released version, whereas this README is for the last development version which
may include additions/changes.**
---
Platform support:
| Adapter | OS | Status |
| --------------------- | ---------------| -------------------------------------------------------------|
| inotify | Linux 2.6.32+ | Supported |
| kqueue | BSD, macOS | Supported |
| ReadDirectoryChangesW | Windows | Supported |
| FSEvents | macOS | [Planned](https://github.com/fsnotify/fsnotify/issues/11) |
| FEN | Solaris 11 | [In Progress](https://github.com/fsnotify/fsnotify/pull/371) |
| fanotify | Linux 5.9+ | [Maybe](https://github.com/fsnotify/fsnotify/issues/114) |
| USN Journals | Windows | [Maybe](https://github.com/fsnotify/fsnotify/issues/53) |
| Polling | *All* | [Maybe](https://github.com/fsnotify/fsnotify/issues/9) |
| Backend | OS | Status |
| :-------------------- | :--------- | :------------------------------------------------------------------------ |
| inotify | Linux | Supported |
| kqueue | BSD, macOS | Supported |
| ReadDirectoryChangesW | Windows | Supported |
| FEN | illumos | Supported |
| fanotify | Linux 5.9+ | [Not yet](https://github.com/fsnotify/fsnotify/issues/114) |
| AHAFS | AIX | [aix branch]; experimental due to lack of maintainer and test environment |
| FSEvents | macOS | [Needs support in x/sys/unix][fsevents] |
| USN Journals | Windows | [Needs support in x/sys/windows][usn] |
| Polling | *All* | [Not yet](https://github.com/fsnotify/fsnotify/issues/9) |
Linux and macOS should include Android and iOS, but these are currently untested.
Linux and illumos should include Android and Solaris, but these are currently
untested.
[fsevents]: https://github.com/fsnotify/fsnotify/issues/11#issuecomment-1279133120
[usn]: https://github.com/fsnotify/fsnotify/issues/53#issuecomment-1279829847
[aix branch]: https://github.com/fsnotify/fsnotify/issues/353#issuecomment-1284590129
Usage
-----
@ -83,20 +85,23 @@ run with:
% go run ./cmd/fsnotify
Further detailed documentation can be found in godoc:
https://pkg.go.dev/github.com/fsnotify/fsnotify
FAQ
---
### Will a file still be watched when it's moved to another directory?
No, not unless you are watching the location it was moved to.
### Are subdirectories watched too?
### Are subdirectories watched?
No, you must add watches for any directory you want to watch (a recursive
watcher is on the roadmap: [#18]).
[#18]: https://github.com/fsnotify/fsnotify/issues/18
### Do I have to watch the Error and Event channels in a goroutine?
As of now, yes (you can read both channels in the same goroutine using `select`,
you don't need a separate goroutine for both channels; see the example).
Yes. You can read both channels in the same goroutine using `select` (you don't
need a separate goroutine for both channels; see the example).
### Why don't notifications work with NFS, SMB, FUSE, /proc, or /sys?
fsnotify requires support from underlying OS to work. The current NFS and SMB
@ -107,6 +112,32 @@ This could be fixed with a polling watcher ([#9]), but it's not yet implemented.
[#9]: https://github.com/fsnotify/fsnotify/issues/9
### Why do I get many Chmod events?
Some programs may generate a lot of attribute changes; for example Spotlight on
macOS, anti-virus programs, backup applications, and some others are known to do
this. As a rule, it's typically best to ignore Chmod events. They're often not
useful, and tend to cause problems.
Spotlight indexing on macOS can result in multiple events (see [#15]). A
temporary workaround is to add your folder(s) to the *Spotlight Privacy
settings* until we have a native FSEvents implementation (see [#11]).
[#11]: https://github.com/fsnotify/fsnotify/issues/11
[#15]: https://github.com/fsnotify/fsnotify/issues/15
### Watching a file doesn't work well
Watching individual files (rather than directories) is generally not recommended
as many programs (especially editors) update files atomically: it will write to
a temporary file which is then moved to to destination, overwriting the original
(or some variant thereof). The watcher on the original file is now lost, as that
no longer exists.
The upshot of this is that a power failure or crash won't leave a half-written
file.
Watch the parent directory and use `Event.Name` to filter out files you're not
interested in. There is an example of this in `cmd/fsnotify/file.go`.
Platform-specific notes
-----------------------
### Linux
@ -151,11 +182,3 @@ these platforms.
The sysctl variables `kern.maxfiles` and `kern.maxfilesperproc` can be used to
control the maximum number of open files.
### macOS
Spotlight indexing on macOS can result in multiple events (see [#15]). A temporary
workaround is to add your folder(s) to the *Spotlight Privacy settings* until we
have a native FSEvents implementation (see [#11]).
[#11]: https://github.com/fsnotify/fsnotify/issues/11
[#15]: https://github.com/fsnotify/fsnotify/issues/15

View file

@ -1,162 +1,484 @@
//go:build solaris
// +build solaris
// FEN backend for illumos (supported) and Solaris (untested, but should work).
//
// See port_create(3c) etc. for docs. https://www.illumos.org/man/3C/port_create
package fsnotify
import (
"errors"
"fmt"
"os"
"path/filepath"
"sync"
"time"
"github.com/fsnotify/fsnotify/internal"
"golang.org/x/sys/unix"
)
// Watcher watches a set of paths, delivering events on a channel.
//
// A watcher should not be copied (e.g. pass it by pointer, rather than by
// value).
//
// # Linux notes
//
// When a file is removed a Remove event won't be emitted until all file
// descriptors are closed, and deletes will always emit a Chmod. For example:
//
// fp := os.Open("file")
// os.Remove("file") // Triggers Chmod
// fp.Close() // Triggers Remove
//
// This is the event that inotify sends, so not much can be changed about this.
//
// The fs.inotify.max_user_watches sysctl variable specifies the upper limit
// for the number of watches per user, and fs.inotify.max_user_instances
// specifies the maximum number of inotify instances per user. Every Watcher you
// create is an "instance", and every path you add is a "watch".
//
// These are also exposed in /proc as /proc/sys/fs/inotify/max_user_watches and
// /proc/sys/fs/inotify/max_user_instances
//
// To increase them you can use sysctl or write the value to the /proc file:
//
// # Default values on Linux 5.18
// sysctl fs.inotify.max_user_watches=124983
// sysctl fs.inotify.max_user_instances=128
//
// To make the changes persist on reboot edit /etc/sysctl.conf or
// /usr/lib/sysctl.d/50-default.conf (details differ per Linux distro; check
// your distro's documentation):
//
// fs.inotify.max_user_watches=124983
// fs.inotify.max_user_instances=128
//
// Reaching the limit will result in a "no space left on device" or "too many open
// files" error.
//
// # kqueue notes (macOS, BSD)
//
// kqueue requires opening a file descriptor for every file that's being watched;
// so if you're watching a directory with five files then that's six file
// descriptors. You will run in to your system's "max open files" limit faster on
// these platforms.
//
// The sysctl variables kern.maxfiles and kern.maxfilesperproc can be used to
// control the maximum number of open files, as well as /etc/login.conf on BSD
// systems.
//
// # macOS notes
//
// Spotlight indexing on macOS can result in multiple events (see [#15]). A
// temporary workaround is to add your folder(s) to the "Spotlight Privacy
// Settings" until we have a native FSEvents implementation (see [#11]).
//
// [#11]: https://github.com/fsnotify/fsnotify/issues/11
// [#15]: https://github.com/fsnotify/fsnotify/issues/15
type Watcher struct {
// Events sends the filesystem change events.
//
// fsnotify can send the following events; a "path" here can refer to a
// file, directory, symbolic link, or special file like a FIFO.
//
// fsnotify.Create A new path was created; this may be followed by one
// or more Write events if data also gets written to a
// file.
//
// fsnotify.Remove A path was removed.
//
// fsnotify.Rename A path was renamed. A rename is always sent with the
// old path as Event.Name, and a Create event will be
// sent with the new name. Renames are only sent for
// paths that are currently watched; e.g. moving an
// unmonitored file into a monitored directory will
// show up as just a Create. Similarly, renaming a file
// to outside a monitored directory will show up as
// only a Rename.
//
// fsnotify.Write A file or named pipe was written to. A Truncate will
// also trigger a Write. A single "write action"
// initiated by the user may show up as one or multiple
// writes, depending on when the system syncs things to
// disk. For example when compiling a large Go program
// you may get hundreds of Write events, so you
// probably want to wait until you've stopped receiving
// them (see the dedup example in cmd/fsnotify).
//
// fsnotify.Chmod Attributes were changed. On Linux this is also sent
// when a file is removed (or more accurately, when a
// link to an inode is removed). On kqueue it's sent
// and on kqueue when a file is truncated. On Windows
// it's never sent.
type fen struct {
Events chan Event
// Errors sends any errors.
Errors chan error
mu sync.Mutex
port *unix.EventPort
done chan struct{} // Channel for sending a "quit message" to the reader goroutine
dirs map[string]Op // Explicitly watched directories
watches map[string]Op // Explicitly watched non-directories
}
// NewWatcher creates a new Watcher.
func NewWatcher() (*Watcher, error) {
return nil, errors.New("FEN based watcher not yet supported for fsnotify\n")
func newBackend(ev chan Event, errs chan error) (backend, error) {
return newBufferedBackend(0, ev, errs)
}
// Close removes all watches and closes the events channel.
func (w *Watcher) Close() error {
func newBufferedBackend(sz uint, ev chan Event, errs chan error) (backend, error) {
w := &fen{
Events: ev,
Errors: errs,
dirs: make(map[string]Op),
watches: make(map[string]Op),
done: make(chan struct{}),
}
var err error
w.port, err = unix.NewEventPort()
if err != nil {
return nil, fmt.Errorf("fsnotify.NewWatcher: %w", err)
}
go w.readEvents()
return w, nil
}
// sendEvent attempts to send an event to the user, returning true if the event
// was put in the channel successfully and false if the watcher has been closed.
func (w *fen) sendEvent(name string, op Op) (sent bool) {
select {
case <-w.done:
return false
case w.Events <- Event{Name: name, Op: op}:
return true
}
}
// sendError attempts to send an error to the user, returning true if the error
// was put in the channel successfully and false if the watcher has been closed.
func (w *fen) sendError(err error) (sent bool) {
if err == nil {
return true
}
select {
case <-w.done:
return false
case w.Errors <- err:
return true
}
}
func (w *fen) isClosed() bool {
select {
case <-w.done:
return true
default:
return false
}
}
func (w *fen) Close() error {
// Take the lock used by associateFile to prevent lingering events from
// being processed after the close
w.mu.Lock()
defer w.mu.Unlock()
if w.isClosed() {
return nil
}
close(w.done)
return w.port.Close()
}
func (w *fen) Add(name string) error { return w.AddWith(name) }
func (w *fen) AddWith(name string, opts ...addOpt) error {
if w.isClosed() {
return ErrClosed
}
if debug {
fmt.Fprintf(os.Stderr, "FSNOTIFY_DEBUG: %s AddWith(%q)\n",
time.Now().Format("15:04:05.000000000"), name)
}
with := getOptions(opts...)
if !w.xSupports(with.op) {
return fmt.Errorf("%w: %s", xErrUnsupported, with.op)
}
// Currently we resolve symlinks that were explicitly requested to be
// watched. Otherwise we would use LStat here.
stat, err := os.Stat(name)
if err != nil {
return err
}
// Associate all files in the directory.
if stat.IsDir() {
err := w.handleDirectory(name, stat, true, w.associateFile)
if err != nil {
return err
}
w.mu.Lock()
w.dirs[name] = with.op
w.mu.Unlock()
return nil
}
err = w.associateFile(name, stat, true)
if err != nil {
return err
}
w.mu.Lock()
w.watches[name] = with.op
w.mu.Unlock()
return nil
}
// Add starts monitoring the path for changes.
//
// A path can only be watched once; attempting to watch it more than once will
// return an error. Paths that do not yet exist on the filesystem cannot be
// added. A watch will be automatically removed if the path is deleted.
//
// A path will remain watched if it gets renamed to somewhere else on the same
// filesystem, but the monitor will get removed if the path gets deleted and
// re-created, or if it's moved to a different filesystem.
//
// Notifications on network filesystems (NFS, SMB, FUSE, etc.) or special
// filesystems (/proc, /sys, etc.) generally don't work.
//
// # Watching directories
//
// All files in a directory are monitored, including new files that are created
// after the watcher is started. Subdirectories are not watched (i.e. it's
// non-recursive).
//
// # Watching files
//
// Watching individual files (rather than directories) is generally not
// recommended as many tools update files atomically. Instead of "just" writing
// to the file a temporary file will be written to first, and if successful the
// temporary file is moved to to destination removing the original, or some
// variant thereof. The watcher on the original file is now lost, as it no
// longer exists.
//
// Instead, watch the parent directory and use Event.Name to filter out files
// you're not interested in. There is an example of this in [cmd/fsnotify/file.go].
func (w *Watcher) Add(name string) error {
func (w *fen) Remove(name string) error {
if w.isClosed() {
return nil
}
if !w.port.PathIsWatched(name) {
return fmt.Errorf("%w: %s", ErrNonExistentWatch, name)
}
if debug {
fmt.Fprintf(os.Stderr, "FSNOTIFY_DEBUG: %s Remove(%q)\n",
time.Now().Format("15:04:05.000000000"), name)
}
// The user has expressed an intent. Immediately remove this name from
// whichever watch list it might be in. If it's not in there the delete
// doesn't cause harm.
w.mu.Lock()
delete(w.watches, name)
delete(w.dirs, name)
w.mu.Unlock()
stat, err := os.Stat(name)
if err != nil {
return err
}
// Remove associations for every file in the directory.
if stat.IsDir() {
err := w.handleDirectory(name, stat, false, w.dissociateFile)
if err != nil {
return err
}
return nil
}
err = w.port.DissociatePath(name)
if err != nil {
return err
}
return nil
}
// Remove stops monitoring the path for changes.
//
// Directories are always removed non-recursively. For example, if you added
// /tmp/dir and /tmp/dir/subdir then you will need to remove both.
//
// Removing a path that has not yet been added returns [ErrNonExistentWatch].
func (w *Watcher) Remove(name string) error {
// readEvents contains the main loop that runs in a goroutine watching for events.
func (w *fen) readEvents() {
// If this function returns, the watcher has been closed and we can close
// these channels
defer func() {
close(w.Errors)
close(w.Events)
}()
pevents := make([]unix.PortEvent, 8)
for {
count, err := w.port.Get(pevents, 1, nil)
if err != nil && err != unix.ETIME {
// Interrupted system call (count should be 0) ignore and continue
if errors.Is(err, unix.EINTR) && count == 0 {
continue
}
// Get failed because we called w.Close()
if errors.Is(err, unix.EBADF) && w.isClosed() {
return
}
// There was an error not caused by calling w.Close()
if !w.sendError(err) {
return
}
}
p := pevents[:count]
for _, pevent := range p {
if pevent.Source != unix.PORT_SOURCE_FILE {
// Event from unexpected source received; should never happen.
if !w.sendError(errors.New("Event from unexpected source received")) {
return
}
continue
}
if debug {
internal.Debug(pevent.Path, pevent.Events)
}
err = w.handleEvent(&pevent)
if !w.sendError(err) {
return
}
}
}
}
func (w *fen) handleDirectory(path string, stat os.FileInfo, follow bool, handler func(string, os.FileInfo, bool) error) error {
files, err := os.ReadDir(path)
if err != nil {
return err
}
// Handle all children of the directory.
for _, entry := range files {
finfo, err := entry.Info()
if err != nil {
return err
}
err = handler(filepath.Join(path, finfo.Name()), finfo, false)
if err != nil {
return err
}
}
// And finally handle the directory itself.
return handler(path, stat, follow)
}
// handleEvent might need to emit more than one fsnotify event if the events
// bitmap matches more than one event type (e.g. the file was both modified and
// had the attributes changed between when the association was created and the
// when event was returned)
func (w *fen) handleEvent(event *unix.PortEvent) error {
var (
events = event.Events
path = event.Path
fmode = event.Cookie.(os.FileMode)
reRegister = true
)
w.mu.Lock()
_, watchedDir := w.dirs[path]
_, watchedPath := w.watches[path]
w.mu.Unlock()
isWatched := watchedDir || watchedPath
if events&unix.FILE_DELETE != 0 {
if !w.sendEvent(path, Remove) {
return nil
}
reRegister = false
}
if events&unix.FILE_RENAME_FROM != 0 {
if !w.sendEvent(path, Rename) {
return nil
}
// Don't keep watching the new file name
reRegister = false
}
if events&unix.FILE_RENAME_TO != 0 {
// We don't report a Rename event for this case, because Rename events
// are interpreted as referring to the _old_ name of the file, and in
// this case the event would refer to the new name of the file. This
// type of rename event is not supported by fsnotify.
// inotify reports a Remove event in this case, so we simulate this
// here.
if !w.sendEvent(path, Remove) {
return nil
}
// Don't keep watching the file that was removed
reRegister = false
}
// The file is gone, nothing left to do.
if !reRegister {
if watchedDir {
w.mu.Lock()
delete(w.dirs, path)
w.mu.Unlock()
}
if watchedPath {
w.mu.Lock()
delete(w.watches, path)
w.mu.Unlock()
}
return nil
}
// If we didn't get a deletion the file still exists and we're going to have
// to watch it again. Let's Stat it now so that we can compare permissions
// and have what we need to continue watching the file
stat, err := os.Lstat(path)
if err != nil {
// This is unexpected, but we should still emit an event. This happens
// most often on "rm -r" of a subdirectory inside a watched directory We
// get a modify event of something happening inside, but by the time we
// get here, the sudirectory is already gone. Clearly we were watching
// this path but now it is gone. Let's tell the user that it was
// removed.
if !w.sendEvent(path, Remove) {
return nil
}
// Suppress extra write events on removed directories; they are not
// informative and can be confusing.
return nil
}
// resolve symlinks that were explicitly watched as we would have at Add()
// time. this helps suppress spurious Chmod events on watched symlinks
if isWatched {
stat, err = os.Stat(path)
if err != nil {
// The symlink still exists, but the target is gone. Report the
// Remove similar to above.
if !w.sendEvent(path, Remove) {
return nil
}
// Don't return the error
}
}
if events&unix.FILE_MODIFIED != 0 {
if fmode.IsDir() && watchedDir {
if err := w.updateDirectory(path); err != nil {
return err
}
} else {
if !w.sendEvent(path, Write) {
return nil
}
}
}
if events&unix.FILE_ATTRIB != 0 && stat != nil {
// Only send Chmod if perms changed
if stat.Mode().Perm() != fmode.Perm() {
if !w.sendEvent(path, Chmod) {
return nil
}
}
}
if stat != nil {
// If we get here, it means we've hit an event above that requires us to
// continue watching the file or directory
return w.associateFile(path, stat, isWatched)
}
return nil
}
func (w *fen) updateDirectory(path string) error {
// The directory was modified, so we must find unwatched entities and watch
// them. If something was removed from the directory, nothing will happen,
// as everything else should still be watched.
files, err := os.ReadDir(path)
if err != nil {
return err
}
for _, entry := range files {
path := filepath.Join(path, entry.Name())
if w.port.PathIsWatched(path) {
continue
}
finfo, err := entry.Info()
if err != nil {
return err
}
err = w.associateFile(path, finfo, false)
if !w.sendError(err) {
return nil
}
if !w.sendEvent(path, Create) {
return nil
}
}
return nil
}
func (w *fen) associateFile(path string, stat os.FileInfo, follow bool) error {
if w.isClosed() {
return ErrClosed
}
// This is primarily protecting the call to AssociatePath but it is
// important and intentional that the call to PathIsWatched is also
// protected by this mutex. Without this mutex, AssociatePath has been seen
// to error out that the path is already associated.
w.mu.Lock()
defer w.mu.Unlock()
if w.port.PathIsWatched(path) {
// Remove the old association in favor of this one If we get ENOENT,
// then while the x/sys/unix wrapper still thought that this path was
// associated, the underlying event port did not. This call will have
// cleared up that discrepancy. The most likely cause is that the event
// has fired but we haven't processed it yet.
err := w.port.DissociatePath(path)
if err != nil && !errors.Is(err, unix.ENOENT) {
return err
}
}
var events int
if !follow {
// Watch symlinks themselves rather than their targets unless this entry
// is explicitly watched.
events |= unix.FILE_NOFOLLOW
}
if true { // TODO: implement withOps()
events |= unix.FILE_MODIFIED
}
if true {
events |= unix.FILE_ATTRIB
}
return w.port.AssociatePath(path, stat, events, stat.Mode())
}
func (w *fen) dissociateFile(path string, stat os.FileInfo, unused bool) error {
if !w.port.PathIsWatched(path) {
return nil
}
return w.port.DissociatePath(path)
}
func (w *fen) WatchList() []string {
if w.isClosed() {
return nil
}
w.mu.Lock()
defer w.mu.Unlock()
entries := make([]string, 0, len(w.watches)+len(w.dirs))
for pathname := range w.dirs {
entries = append(entries, pathname)
}
for pathname := range w.watches {
entries = append(entries, pathname)
}
return entries
}
func (w *fen) xSupports(op Op) bool {
if op.Has(xUnportableOpen) || op.Has(xUnportableRead) ||
op.Has(xUnportableCloseWrite) || op.Has(xUnportableCloseRead) {
return false
}
return true
}

View file

@ -1,5 +1,4 @@
//go:build linux
// +build linux
//go:build linux && !appengine
package fsnotify
@ -7,142 +6,188 @@ import (
"errors"
"fmt"
"io"
"io/fs"
"os"
"path/filepath"
"strings"
"sync"
"time"
"unsafe"
"github.com/fsnotify/fsnotify/internal"
"golang.org/x/sys/unix"
)
// Watcher watches a set of paths, delivering events on a channel.
//
// A watcher should not be copied (e.g. pass it by pointer, rather than by
// value).
//
// # Linux notes
//
// When a file is removed a Remove event won't be emitted until all file
// descriptors are closed, and deletes will always emit a Chmod. For example:
//
// fp := os.Open("file")
// os.Remove("file") // Triggers Chmod
// fp.Close() // Triggers Remove
//
// This is the event that inotify sends, so not much can be changed about this.
//
// The fs.inotify.max_user_watches sysctl variable specifies the upper limit
// for the number of watches per user, and fs.inotify.max_user_instances
// specifies the maximum number of inotify instances per user. Every Watcher you
// create is an "instance", and every path you add is a "watch".
//
// These are also exposed in /proc as /proc/sys/fs/inotify/max_user_watches and
// /proc/sys/fs/inotify/max_user_instances
//
// To increase them you can use sysctl or write the value to the /proc file:
//
// # Default values on Linux 5.18
// sysctl fs.inotify.max_user_watches=124983
// sysctl fs.inotify.max_user_instances=128
//
// To make the changes persist on reboot edit /etc/sysctl.conf or
// /usr/lib/sysctl.d/50-default.conf (details differ per Linux distro; check
// your distro's documentation):
//
// fs.inotify.max_user_watches=124983
// fs.inotify.max_user_instances=128
//
// Reaching the limit will result in a "no space left on device" or "too many open
// files" error.
//
// # kqueue notes (macOS, BSD)
//
// kqueue requires opening a file descriptor for every file that's being watched;
// so if you're watching a directory with five files then that's six file
// descriptors. You will run in to your system's "max open files" limit faster on
// these platforms.
//
// The sysctl variables kern.maxfiles and kern.maxfilesperproc can be used to
// control the maximum number of open files, as well as /etc/login.conf on BSD
// systems.
//
// # macOS notes
//
// Spotlight indexing on macOS can result in multiple events (see [#15]). A
// temporary workaround is to add your folder(s) to the "Spotlight Privacy
// Settings" until we have a native FSEvents implementation (see [#11]).
//
// [#11]: https://github.com/fsnotify/fsnotify/issues/11
// [#15]: https://github.com/fsnotify/fsnotify/issues/15
type Watcher struct {
// Events sends the filesystem change events.
//
// fsnotify can send the following events; a "path" here can refer to a
// file, directory, symbolic link, or special file like a FIFO.
//
// fsnotify.Create A new path was created; this may be followed by one
// or more Write events if data also gets written to a
// file.
//
// fsnotify.Remove A path was removed.
//
// fsnotify.Rename A path was renamed. A rename is always sent with the
// old path as Event.Name, and a Create event will be
// sent with the new name. Renames are only sent for
// paths that are currently watched; e.g. moving an
// unmonitored file into a monitored directory will
// show up as just a Create. Similarly, renaming a file
// to outside a monitored directory will show up as
// only a Rename.
//
// fsnotify.Write A file or named pipe was written to. A Truncate will
// also trigger a Write. A single "write action"
// initiated by the user may show up as one or multiple
// writes, depending on when the system syncs things to
// disk. For example when compiling a large Go program
// you may get hundreds of Write events, so you
// probably want to wait until you've stopped receiving
// them (see the dedup example in cmd/fsnotify).
//
// fsnotify.Chmod Attributes were changed. On Linux this is also sent
// when a file is removed (or more accurately, when a
// link to an inode is removed). On kqueue it's sent
// and on kqueue when a file is truncated. On Windows
// it's never sent.
type inotify struct {
Events chan Event
// Errors sends any errors.
Errors chan error
// Store fd here as os.File.Read() will no longer return on close after
// calling Fd(). See: https://github.com/golang/go/issues/26439
fd int
mu sync.Mutex // Map access
inotifyFile *os.File
watches map[string]*watch // Map of inotify watches (key: path)
paths map[int]string // Map of watched paths (key: watch descriptor)
done chan struct{} // Channel for sending a "quit message" to the reader goroutine
doneResp chan struct{} // Channel to respond to Close
watches *watches
done chan struct{} // Channel for sending a "quit message" to the reader goroutine
doneMu sync.Mutex
doneResp chan struct{} // Channel to respond to Close
// Store rename cookies in an array, with the index wrapping to 0. Almost
// all of the time what we get is a MOVED_FROM to set the cookie and the
// next event inotify sends will be MOVED_TO to read it. However, this is
// not guaranteed as described in inotify(7) and we may get other events
// between the two MOVED_* events (including other MOVED_* ones).
//
// A second issue is that moving a file outside the watched directory will
// trigger a MOVED_FROM to set the cookie, but we never see the MOVED_TO to
// read and delete it. So just storing it in a map would slowly leak memory.
//
// Doing it like this gives us a simple fast LRU-cache that won't allocate.
// Ten items should be more than enough for our purpose, and a loop over
// such a short array is faster than a map access anyway (not that it hugely
// matters since we're talking about hundreds of ns at the most, but still).
cookies [10]koekje
cookieIndex uint8
cookiesMu sync.Mutex
}
// NewWatcher creates a new Watcher.
func NewWatcher() (*Watcher, error) {
// Create inotify fd
// Need to set the FD to nonblocking mode in order for SetDeadline methods to work
// Otherwise, blocking i/o operations won't terminate on close
type (
watches struct {
mu sync.RWMutex
wd map[uint32]*watch // wd → watch
path map[string]uint32 // pathname → wd
}
watch struct {
wd uint32 // Watch descriptor (as returned by the inotify_add_watch() syscall)
flags uint32 // inotify flags of this watch (see inotify(7) for the list of valid flags)
path string // Watch path.
recurse bool // Recursion with ./...?
}
koekje struct {
cookie uint32
path string
}
)
func newWatches() *watches {
return &watches{
wd: make(map[uint32]*watch),
path: make(map[string]uint32),
}
}
func (w *watches) len() int {
w.mu.RLock()
defer w.mu.RUnlock()
return len(w.wd)
}
func (w *watches) add(ww *watch) {
w.mu.Lock()
defer w.mu.Unlock()
w.wd[ww.wd] = ww
w.path[ww.path] = ww.wd
}
func (w *watches) remove(wd uint32) {
w.mu.Lock()
defer w.mu.Unlock()
watch := w.wd[wd] // Could have had Remove() called. See #616.
if watch == nil {
return
}
delete(w.path, watch.path)
delete(w.wd, wd)
}
func (w *watches) removePath(path string) ([]uint32, error) {
w.mu.Lock()
defer w.mu.Unlock()
path, recurse := recursivePath(path)
wd, ok := w.path[path]
if !ok {
return nil, fmt.Errorf("%w: %s", ErrNonExistentWatch, path)
}
watch := w.wd[wd]
if recurse && !watch.recurse {
return nil, fmt.Errorf("can't use /... with non-recursive watch %q", path)
}
delete(w.path, path)
delete(w.wd, wd)
if !watch.recurse {
return []uint32{wd}, nil
}
wds := make([]uint32, 0, 8)
wds = append(wds, wd)
for p, rwd := range w.path {
if filepath.HasPrefix(p, path) {
delete(w.path, p)
delete(w.wd, rwd)
wds = append(wds, rwd)
}
}
return wds, nil
}
func (w *watches) byPath(path string) *watch {
w.mu.RLock()
defer w.mu.RUnlock()
return w.wd[w.path[path]]
}
func (w *watches) byWd(wd uint32) *watch {
w.mu.RLock()
defer w.mu.RUnlock()
return w.wd[wd]
}
func (w *watches) updatePath(path string, f func(*watch) (*watch, error)) error {
w.mu.Lock()
defer w.mu.Unlock()
var existing *watch
wd, ok := w.path[path]
if ok {
existing = w.wd[wd]
}
upd, err := f(existing)
if err != nil {
return err
}
if upd != nil {
w.wd[upd.wd] = upd
w.path[upd.path] = upd.wd
if upd.wd != wd {
delete(w.wd, wd)
}
}
return nil
}
func newBackend(ev chan Event, errs chan error) (backend, error) {
return newBufferedBackend(0, ev, errs)
}
func newBufferedBackend(sz uint, ev chan Event, errs chan error) (backend, error) {
// Need to set nonblocking mode for SetDeadline to work, otherwise blocking
// I/O operations won't terminate on close.
fd, errno := unix.InotifyInit1(unix.IN_CLOEXEC | unix.IN_NONBLOCK)
if fd == -1 {
return nil, errno
}
w := &Watcher{
w := &inotify{
Events: ev,
Errors: errs,
fd: fd,
inotifyFile: os.NewFile(uintptr(fd), ""),
watches: make(map[string]*watch),
paths: make(map[int]string),
Events: make(chan Event),
Errors: make(chan error),
watches: newWatches(),
done: make(chan struct{}),
doneResp: make(chan struct{}),
}
@ -152,26 +197,29 @@ func NewWatcher() (*Watcher, error) {
}
// Returns true if the event was sent, or false if watcher is closed.
func (w *Watcher) sendEvent(e Event) bool {
func (w *inotify) sendEvent(e Event) bool {
select {
case <-w.done:
return false
case w.Events <- e:
return true
case <-w.done:
}
return false
}
// Returns true if the error was sent, or false if watcher is closed.
func (w *Watcher) sendError(err error) bool {
select {
case w.Errors <- err:
func (w *inotify) sendError(err error) bool {
if err == nil {
return true
}
select {
case <-w.done:
return false
case w.Errors <- err:
return true
}
}
func (w *Watcher) isClosed() bool {
func (w *inotify) isClosed() bool {
select {
case <-w.done:
return true
@ -180,17 +228,14 @@ func (w *Watcher) isClosed() bool {
}
}
// Close removes all watches and closes the events channel.
func (w *Watcher) Close() error {
w.mu.Lock()
func (w *inotify) Close() error {
w.doneMu.Lock()
if w.isClosed() {
w.mu.Unlock()
w.doneMu.Unlock()
return nil
}
// Send 'close' signal to goroutine, and set the Watcher to closed.
close(w.done)
w.mu.Unlock()
w.doneMu.Unlock()
// Causes any blocking reads to return with an error, provided the file
// still supports deadline operations.
@ -205,138 +250,168 @@ func (w *Watcher) Close() error {
return nil
}
// Add starts monitoring the path for changes.
//
// A path can only be watched once; attempting to watch it more than once will
// return an error. Paths that do not yet exist on the filesystem cannot be
// added. A watch will be automatically removed if the path is deleted.
//
// A path will remain watched if it gets renamed to somewhere else on the same
// filesystem, but the monitor will get removed if the path gets deleted and
// re-created, or if it's moved to a different filesystem.
//
// Notifications on network filesystems (NFS, SMB, FUSE, etc.) or special
// filesystems (/proc, /sys, etc.) generally don't work.
//
// # Watching directories
//
// All files in a directory are monitored, including new files that are created
// after the watcher is started. Subdirectories are not watched (i.e. it's
// non-recursive).
//
// # Watching files
//
// Watching individual files (rather than directories) is generally not
// recommended as many tools update files atomically. Instead of "just" writing
// to the file a temporary file will be written to first, and if successful the
// temporary file is moved to to destination removing the original, or some
// variant thereof. The watcher on the original file is now lost, as it no
// longer exists.
//
// Instead, watch the parent directory and use Event.Name to filter out files
// you're not interested in. There is an example of this in [cmd/fsnotify/file.go].
func (w *Watcher) Add(name string) error {
name = filepath.Clean(name)
func (w *inotify) Add(name string) error { return w.AddWith(name) }
func (w *inotify) AddWith(path string, opts ...addOpt) error {
if w.isClosed() {
return errors.New("inotify instance already closed")
return ErrClosed
}
if debug {
fmt.Fprintf(os.Stderr, "FSNOTIFY_DEBUG: %s AddWith(%q)\n",
time.Now().Format("15:04:05.000000000"), path)
}
var flags uint32 = unix.IN_MOVED_TO | unix.IN_MOVED_FROM |
unix.IN_CREATE | unix.IN_ATTRIB | unix.IN_MODIFY |
unix.IN_MOVE_SELF | unix.IN_DELETE | unix.IN_DELETE_SELF
w.mu.Lock()
defer w.mu.Unlock()
watchEntry := w.watches[name]
if watchEntry != nil {
flags |= watchEntry.flags | unix.IN_MASK_ADD
}
wd, errno := unix.InotifyAddWatch(w.fd, name, flags)
if wd == -1 {
return errno
with := getOptions(opts...)
if !w.xSupports(with.op) {
return fmt.Errorf("%w: %s", xErrUnsupported, with.op)
}
if watchEntry == nil {
w.watches[name] = &watch{wd: uint32(wd), flags: flags}
w.paths[wd] = name
} else {
watchEntry.wd = uint32(wd)
watchEntry.flags = flags
path, recurse := recursivePath(path)
if recurse {
return filepath.WalkDir(path, func(root string, d fs.DirEntry, err error) error {
if err != nil {
return err
}
if !d.IsDir() {
if root == path {
return fmt.Errorf("fsnotify: not a directory: %q", path)
}
return nil
}
// Send a Create event when adding new directory from a recursive
// watch; this is for "mkdir -p one/two/three". Usually all those
// directories will be created before we can set up watchers on the
// subdirectories, so only "one" would be sent as a Create event and
// not "one/two" and "one/two/three" (inotifywait -r has the same
// problem).
if with.sendCreate && root != path {
w.sendEvent(Event{Name: root, Op: Create})
}
return w.add(root, with, true)
})
}
return w.add(path, with, false)
}
func (w *inotify) add(path string, with withOpts, recurse bool) error {
var flags uint32
if with.noFollow {
flags |= unix.IN_DONT_FOLLOW
}
if with.op.Has(Create) {
flags |= unix.IN_CREATE
}
if with.op.Has(Write) {
flags |= unix.IN_MODIFY
}
if with.op.Has(Remove) {
flags |= unix.IN_DELETE | unix.IN_DELETE_SELF
}
if with.op.Has(Rename) {
flags |= unix.IN_MOVED_TO | unix.IN_MOVED_FROM | unix.IN_MOVE_SELF
}
if with.op.Has(Chmod) {
flags |= unix.IN_ATTRIB
}
if with.op.Has(xUnportableOpen) {
flags |= unix.IN_OPEN
}
if with.op.Has(xUnportableRead) {
flags |= unix.IN_ACCESS
}
if with.op.Has(xUnportableCloseWrite) {
flags |= unix.IN_CLOSE_WRITE
}
if with.op.Has(xUnportableCloseRead) {
flags |= unix.IN_CLOSE_NOWRITE
}
return w.register(path, flags, recurse)
}
func (w *inotify) register(path string, flags uint32, recurse bool) error {
return w.watches.updatePath(path, func(existing *watch) (*watch, error) {
if existing != nil {
flags |= existing.flags | unix.IN_MASK_ADD
}
wd, err := unix.InotifyAddWatch(w.fd, path, flags)
if wd == -1 {
return nil, err
}
if existing == nil {
return &watch{
wd: uint32(wd),
path: path,
flags: flags,
recurse: recurse,
}, nil
}
existing.wd = uint32(wd)
existing.flags = flags
return existing, nil
})
}
func (w *inotify) Remove(name string) error {
if w.isClosed() {
return nil
}
if debug {
fmt.Fprintf(os.Stderr, "FSNOTIFY_DEBUG: %s Remove(%q)\n",
time.Now().Format("15:04:05.000000000"), name)
}
return w.remove(filepath.Clean(name))
}
func (w *inotify) remove(name string) error {
wds, err := w.watches.removePath(name)
if err != nil {
return err
}
for _, wd := range wds {
_, err := unix.InotifyRmWatch(w.fd, wd)
if err != nil {
// TODO: Perhaps it's not helpful to return an error here in every
// case; the only two possible errors are:
//
// EBADF, which happens when w.fd is not a valid file descriptor of
// any kind.
//
// EINVAL, which is when fd is not an inotify descriptor or wd is
// not a valid watch descriptor. Watch descriptors are invalidated
// when they are removed explicitly or implicitly; explicitly by
// inotify_rm_watch, implicitly when the file they are watching is
// deleted.
return err
}
}
return nil
}
// Remove stops monitoring the path for changes.
//
// Directories are always removed non-recursively. For example, if you added
// /tmp/dir and /tmp/dir/subdir then you will need to remove both.
//
// Removing a path that has not yet been added returns [ErrNonExistentWatch].
func (w *Watcher) Remove(name string) error {
name = filepath.Clean(name)
// Fetch the watch.
w.mu.Lock()
defer w.mu.Unlock()
watch, ok := w.watches[name]
// Remove it from inotify.
if !ok {
return fmt.Errorf("%w: %s", ErrNonExistentWatch, name)
func (w *inotify) WatchList() []string {
if w.isClosed() {
return nil
}
// We successfully removed the watch if InotifyRmWatch doesn't return an
// error, we need to clean up our internal state to ensure it matches
// inotify's kernel state.
delete(w.paths, int(watch.wd))
delete(w.watches, name)
// inotify_rm_watch will return EINVAL if the file has been deleted;
// the inotify will already have been removed.
// watches and pathes are deleted in ignoreLinux() implicitly and asynchronously
// by calling inotify_rm_watch() below. e.g. readEvents() goroutine receives IN_IGNORE
// so that EINVAL means that the wd is being rm_watch()ed or its file removed
// by another thread and we have not received IN_IGNORE event.
success, errno := unix.InotifyRmWatch(w.fd, watch.wd)
if success == -1 {
// TODO: Perhaps it's not helpful to return an error here in every case;
// The only two possible errors are:
//
// - EBADF, which happens when w.fd is not a valid file descriptor
// of any kind.
// - EINVAL, which is when fd is not an inotify descriptor or wd
// is not a valid watch descriptor. Watch descriptors are
// invalidated when they are removed explicitly or implicitly;
// explicitly by inotify_rm_watch, implicitly when the file they
// are watching is deleted.
return errno
}
return nil
}
// WatchList returns all paths added with [Add] (and are not yet removed).
func (w *Watcher) WatchList() []string {
w.mu.Lock()
defer w.mu.Unlock()
entries := make([]string, 0, len(w.watches))
for pathname := range w.watches {
entries := make([]string, 0, w.watches.len())
w.watches.mu.RLock()
for pathname := range w.watches.path {
entries = append(entries, pathname)
}
w.watches.mu.RUnlock()
return entries
}
type watch struct {
wd uint32 // Watch descriptor (as returned by the inotify_add_watch() syscall)
flags uint32 // inotify flags of this watch (see inotify(7) for the list of valid flags)
}
// readEvents reads from the inotify file descriptor, converts the
// received events into Event objects and sends them via the Events channel
func (w *Watcher) readEvents() {
func (w *inotify) readEvents() {
defer func() {
close(w.doneResp)
close(w.Errors)
@ -367,14 +442,11 @@ func (w *Watcher) readEvents() {
if n < unix.SizeofInotifyEvent {
var err error
if n == 0 {
// If EOF is received. This should really never happen.
err = io.EOF
err = io.EOF // If EOF is received. This should really never happen.
} else if n < 0 {
// If an error occurred while reading.
err = errno
err = errno // If an error occurred while reading.
} else {
// Read was too short.
err = errors.New("notify: short read in readEvents()")
err = errors.New("notify: short read in readEvents()") // Read was too short.
}
if !w.sendError(err) {
return
@ -382,15 +454,17 @@ func (w *Watcher) readEvents() {
continue
}
var offset uint32
// We don't know how many events we just read into the buffer
// While the offset points to at least one whole event...
var offset uint32
for offset <= uint32(n-unix.SizeofInotifyEvent) {
var (
// Point "raw" to the event in the buffer
raw = (*unix.InotifyEvent)(unsafe.Pointer(&buf[offset]))
mask = uint32(raw.Mask)
nameLen = uint32(raw.Len)
// Move to the next event in the buffer
next = func() { offset += unix.SizeofInotifyEvent + nameLen }
)
if mask&unix.IN_Q_OVERFLOW != 0 {
@ -399,46 +473,124 @@ func (w *Watcher) readEvents() {
}
}
// If the event happened to the watched directory or the watched file, the kernel
// doesn't append the filename to the event, but we would like to always fill the
// the "Name" field with a valid filename. We retrieve the path of the watch from
// the "paths" map.
w.mu.Lock()
name, ok := w.paths[int(raw.Wd)]
// IN_DELETE_SELF occurs when the file/directory being watched is removed.
// This is a sign to clean up the maps, otherwise we are no longer in sync
// with the inotify kernel state which has already deleted the watch
// automatically.
if ok && mask&unix.IN_DELETE_SELF == unix.IN_DELETE_SELF {
delete(w.paths, int(raw.Wd))
delete(w.watches, name)
/// If the event happened to the watched directory or the watched
/// file, the kernel doesn't append the filename to the event, but
/// we would like to always fill the the "Name" field with a valid
/// filename. We retrieve the path of the watch from the "paths"
/// map.
watch := w.watches.byWd(uint32(raw.Wd))
/// Can be nil if Remove() was called in another goroutine for this
/// path inbetween reading the events from the kernel and reading
/// the internal state. Not much we can do about it, so just skip.
/// See #616.
if watch == nil {
next()
continue
}
w.mu.Unlock()
name := watch.path
if nameLen > 0 {
// Point "bytes" at the first byte of the filename
/// Point "bytes" at the first byte of the filename
bytes := (*[unix.PathMax]byte)(unsafe.Pointer(&buf[offset+unix.SizeofInotifyEvent]))[:nameLen:nameLen]
// The filename is padded with NULL bytes. TrimRight() gets rid of those.
/// The filename is padded with NULL bytes. TrimRight() gets rid of those.
name += "/" + strings.TrimRight(string(bytes[0:nameLen]), "\000")
}
event := w.newEvent(name, mask)
if debug {
internal.Debug(name, raw.Mask, raw.Cookie)
}
// Send the events that are not ignored on the events channel
if mask&unix.IN_IGNORED == 0 {
if !w.sendEvent(event) {
return
if mask&unix.IN_IGNORED != 0 { //&& event.Op != 0
next()
continue
}
// inotify will automatically remove the watch on deletes; just need
// to clean our state here.
if mask&unix.IN_DELETE_SELF == unix.IN_DELETE_SELF {
w.watches.remove(watch.wd)
}
// We can't really update the state when a watched path is moved;
// only IN_MOVE_SELF is sent and not IN_MOVED_{FROM,TO}. So remove
// the watch.
if mask&unix.IN_MOVE_SELF == unix.IN_MOVE_SELF {
if watch.recurse {
next() // Do nothing
continue
}
err := w.remove(watch.path)
if err != nil && !errors.Is(err, ErrNonExistentWatch) {
if !w.sendError(err) {
return
}
}
}
// Move to the next event in the buffer
offset += unix.SizeofInotifyEvent + nameLen
/// Skip if we're watching both this path and the parent; the parent
/// will already send a delete so no need to do it twice.
if mask&unix.IN_DELETE_SELF != 0 {
if _, ok := w.watches.path[filepath.Dir(watch.path)]; ok {
next()
continue
}
}
ev := w.newEvent(name, mask, raw.Cookie)
// Need to update watch path for recurse.
if watch.recurse {
isDir := mask&unix.IN_ISDIR == unix.IN_ISDIR
/// New directory created: set up watch on it.
if isDir && ev.Has(Create) {
err := w.register(ev.Name, watch.flags, true)
if !w.sendError(err) {
return
}
// This was a directory rename, so we need to update all
// the children.
//
// TODO: this is of course pretty slow; we should use a
// better data structure for storing all of this, e.g. store
// children in the watch. I have some code for this in my
// kqueue refactor we can use in the future. For now I'm
// okay with this as it's not publicly available.
// Correctness first, performance second.
if ev.renamedFrom != "" {
w.watches.mu.Lock()
for k, ww := range w.watches.wd {
if k == watch.wd || ww.path == ev.Name {
continue
}
if strings.HasPrefix(ww.path, ev.renamedFrom) {
ww.path = strings.Replace(ww.path, ev.renamedFrom, ev.Name, 1)
w.watches.wd[k] = ww
}
}
w.watches.mu.Unlock()
}
}
}
/// Send the events that are not ignored on the events channel
if !w.sendEvent(ev) {
return
}
next()
}
}
}
// newEvent returns an platform-independent Event based on an inotify mask.
func (w *Watcher) newEvent(name string, mask uint32) Event {
func (w *inotify) isRecursive(path string) bool {
ww := w.watches.byPath(path)
if ww == nil { // path could be a file, so also check the Dir.
ww = w.watches.byPath(filepath.Dir(path))
}
return ww != nil && ww.recurse
}
func (w *inotify) newEvent(name string, mask, cookie uint32) Event {
e := Event{Name: name}
if mask&unix.IN_CREATE == unix.IN_CREATE || mask&unix.IN_MOVED_TO == unix.IN_MOVED_TO {
e.Op |= Create
@ -449,11 +601,58 @@ func (w *Watcher) newEvent(name string, mask uint32) Event {
if mask&unix.IN_MODIFY == unix.IN_MODIFY {
e.Op |= Write
}
if mask&unix.IN_OPEN == unix.IN_OPEN {
e.Op |= xUnportableOpen
}
if mask&unix.IN_ACCESS == unix.IN_ACCESS {
e.Op |= xUnportableRead
}
if mask&unix.IN_CLOSE_WRITE == unix.IN_CLOSE_WRITE {
e.Op |= xUnportableCloseWrite
}
if mask&unix.IN_CLOSE_NOWRITE == unix.IN_CLOSE_NOWRITE {
e.Op |= xUnportableCloseRead
}
if mask&unix.IN_MOVE_SELF == unix.IN_MOVE_SELF || mask&unix.IN_MOVED_FROM == unix.IN_MOVED_FROM {
e.Op |= Rename
}
if mask&unix.IN_ATTRIB == unix.IN_ATTRIB {
e.Op |= Chmod
}
if cookie != 0 {
if mask&unix.IN_MOVED_FROM == unix.IN_MOVED_FROM {
w.cookiesMu.Lock()
w.cookies[w.cookieIndex] = koekje{cookie: cookie, path: e.Name}
w.cookieIndex++
if w.cookieIndex > 9 {
w.cookieIndex = 0
}
w.cookiesMu.Unlock()
} else if mask&unix.IN_MOVED_TO == unix.IN_MOVED_TO {
w.cookiesMu.Lock()
var prev string
for _, c := range w.cookies {
if c.cookie == cookie {
prev = c.path
break
}
}
w.cookiesMu.Unlock()
e.renamedFrom = prev
}
}
return e
}
func (w *inotify) xSupports(op Op) bool {
return true // Supports everything.
}
func (w *inotify) state() {
w.watches.mu.Lock()
defer w.watches.mu.Unlock()
for wd, ww := range w.watches.wd {
fmt.Fprintf(os.Stderr, "%4d: recurse=%t %q\n", wd, ww.recurse, ww.path)
}
}

File diff suppressed because it is too large Load diff

View file

@ -1,66 +1,23 @@
//go:build !darwin && !dragonfly && !freebsd && !openbsd && !linux && !netbsd && !solaris && !windows
// +build !darwin,!dragonfly,!freebsd,!openbsd,!linux,!netbsd,!solaris,!windows
//go:build appengine || (!darwin && !dragonfly && !freebsd && !openbsd && !linux && !netbsd && !solaris && !windows)
package fsnotify
import (
"fmt"
"runtime"
)
import "errors"
// Watcher watches a set of files, delivering events to a channel.
type Watcher struct{}
// NewWatcher creates a new Watcher.
func NewWatcher() (*Watcher, error) {
return nil, fmt.Errorf("fsnotify not supported on %s", runtime.GOOS)
type other struct {
Events chan Event
Errors chan error
}
// Close removes all watches and closes the events channel.
func (w *Watcher) Close() error {
return nil
func newBackend(ev chan Event, errs chan error) (backend, error) {
return nil, errors.New("fsnotify not supported on the current platform")
}
// Add starts monitoring the path for changes.
//
// A path can only be watched once; attempting to watch it more than once will
// return an error. Paths that do not yet exist on the filesystem cannot be
// added. A watch will be automatically removed if the path is deleted.
//
// A path will remain watched if it gets renamed to somewhere else on the same
// filesystem, but the monitor will get removed if the path gets deleted and
// re-created, or if it's moved to a different filesystem.
//
// Notifications on network filesystems (NFS, SMB, FUSE, etc.) or special
// filesystems (/proc, /sys, etc.) generally don't work.
//
// # Watching directories
//
// All files in a directory are monitored, including new files that are created
// after the watcher is started. Subdirectories are not watched (i.e. it's
// non-recursive).
//
// # Watching files
//
// Watching individual files (rather than directories) is generally not
// recommended as many tools update files atomically. Instead of "just" writing
// to the file a temporary file will be written to first, and if successful the
// temporary file is moved to to destination removing the original, or some
// variant thereof. The watcher on the original file is now lost, as it no
// longer exists.
//
// Instead, watch the parent directory and use Event.Name to filter out files
// you're not interested in. There is an example of this in [cmd/fsnotify/file.go].
func (w *Watcher) Add(name string) error {
return nil
}
// Remove stops monitoring the path for changes.
//
// Directories are always removed non-recursively. For example, if you added
// /tmp/dir and /tmp/dir/subdir then you will need to remove both.
//
// Removing a path that has not yet been added returns [ErrNonExistentWatch].
func (w *Watcher) Remove(name string) error {
return nil
func newBufferedBackend(sz uint, ev chan Event, errs chan error) (backend, error) {
return newBackend(ev, errs)
}
func (w *other) Close() error { return nil }
func (w *other) WatchList() []string { return nil }
func (w *other) Add(name string) error { return nil }
func (w *other) AddWith(name string, opts ...addOpt) error { return nil }
func (w *other) Remove(name string) error { return nil }
func (w *other) xSupports(op Op) bool { return false }

View file

@ -1,5 +1,8 @@
//go:build windows
// +build windows
// Windows backend based on ReadDirectoryChangesW()
//
// https://learn.microsoft.com/en-us/windows/win32/api/winbase/nf-winbase-readdirectorychangesw
package fsnotify
@ -12,143 +15,60 @@ import (
"runtime"
"strings"
"sync"
"time"
"unsafe"
"github.com/fsnotify/fsnotify/internal"
"golang.org/x/sys/windows"
)
// Watcher watches a set of paths, delivering events on a channel.
//
// A watcher should not be copied (e.g. pass it by pointer, rather than by
// value).
//
// # Linux notes
//
// When a file is removed a Remove event won't be emitted until all file
// descriptors are closed, and deletes will always emit a Chmod. For example:
//
// fp := os.Open("file")
// os.Remove("file") // Triggers Chmod
// fp.Close() // Triggers Remove
//
// This is the event that inotify sends, so not much can be changed about this.
//
// The fs.inotify.max_user_watches sysctl variable specifies the upper limit
// for the number of watches per user, and fs.inotify.max_user_instances
// specifies the maximum number of inotify instances per user. Every Watcher you
// create is an "instance", and every path you add is a "watch".
//
// These are also exposed in /proc as /proc/sys/fs/inotify/max_user_watches and
// /proc/sys/fs/inotify/max_user_instances
//
// To increase them you can use sysctl or write the value to the /proc file:
//
// # Default values on Linux 5.18
// sysctl fs.inotify.max_user_watches=124983
// sysctl fs.inotify.max_user_instances=128
//
// To make the changes persist on reboot edit /etc/sysctl.conf or
// /usr/lib/sysctl.d/50-default.conf (details differ per Linux distro; check
// your distro's documentation):
//
// fs.inotify.max_user_watches=124983
// fs.inotify.max_user_instances=128
//
// Reaching the limit will result in a "no space left on device" or "too many open
// files" error.
//
// # kqueue notes (macOS, BSD)
//
// kqueue requires opening a file descriptor for every file that's being watched;
// so if you're watching a directory with five files then that's six file
// descriptors. You will run in to your system's "max open files" limit faster on
// these platforms.
//
// The sysctl variables kern.maxfiles and kern.maxfilesperproc can be used to
// control the maximum number of open files, as well as /etc/login.conf on BSD
// systems.
//
// # macOS notes
//
// Spotlight indexing on macOS can result in multiple events (see [#15]). A
// temporary workaround is to add your folder(s) to the "Spotlight Privacy
// Settings" until we have a native FSEvents implementation (see [#11]).
//
// [#11]: https://github.com/fsnotify/fsnotify/issues/11
// [#15]: https://github.com/fsnotify/fsnotify/issues/15
type Watcher struct {
// Events sends the filesystem change events.
//
// fsnotify can send the following events; a "path" here can refer to a
// file, directory, symbolic link, or special file like a FIFO.
//
// fsnotify.Create A new path was created; this may be followed by one
// or more Write events if data also gets written to a
// file.
//
// fsnotify.Remove A path was removed.
//
// fsnotify.Rename A path was renamed. A rename is always sent with the
// old path as Event.Name, and a Create event will be
// sent with the new name. Renames are only sent for
// paths that are currently watched; e.g. moving an
// unmonitored file into a monitored directory will
// show up as just a Create. Similarly, renaming a file
// to outside a monitored directory will show up as
// only a Rename.
//
// fsnotify.Write A file or named pipe was written to. A Truncate will
// also trigger a Write. A single "write action"
// initiated by the user may show up as one or multiple
// writes, depending on when the system syncs things to
// disk. For example when compiling a large Go program
// you may get hundreds of Write events, so you
// probably want to wait until you've stopped receiving
// them (see the dedup example in cmd/fsnotify).
//
// fsnotify.Chmod Attributes were changed. On Linux this is also sent
// when a file is removed (or more accurately, when a
// link to an inode is removed). On kqueue it's sent
// and on kqueue when a file is truncated. On Windows
// it's never sent.
type readDirChangesW struct {
Events chan Event
// Errors sends any errors.
Errors chan error
port windows.Handle // Handle to completion port
input chan *input // Inputs to the reader are sent on this channel
quit chan chan<- error
mu sync.Mutex // Protects access to watches, isClosed
watches watchMap // Map of watches (key: i-number)
isClosed bool // Set to true when Close() is first called
mu sync.Mutex // Protects access to watches, closed
watches watchMap // Map of watches (key: i-number)
closed bool // Set to true when Close() is first called
}
// NewWatcher creates a new Watcher.
func NewWatcher() (*Watcher, error) {
func newBackend(ev chan Event, errs chan error) (backend, error) {
return newBufferedBackend(50, ev, errs)
}
func newBufferedBackend(sz uint, ev chan Event, errs chan error) (backend, error) {
port, err := windows.CreateIoCompletionPort(windows.InvalidHandle, 0, 0, 0)
if err != nil {
return nil, os.NewSyscallError("CreateIoCompletionPort", err)
}
w := &Watcher{
w := &readDirChangesW{
Events: ev,
Errors: errs,
port: port,
watches: make(watchMap),
input: make(chan *input, 1),
Events: make(chan Event, 50),
Errors: make(chan error),
quit: make(chan chan<- error, 1),
}
go w.readEvents()
return w, nil
}
func (w *Watcher) sendEvent(name string, mask uint64) bool {
func (w *readDirChangesW) isClosed() bool {
w.mu.Lock()
defer w.mu.Unlock()
return w.closed
}
func (w *readDirChangesW) sendEvent(name, renamedFrom string, mask uint64) bool {
if mask == 0 {
return false
}
event := w.newEvent(name, uint32(mask))
event.renamedFrom = renamedFrom
select {
case ch := <-w.quit:
w.quit <- ch
@ -158,23 +78,25 @@ func (w *Watcher) sendEvent(name string, mask uint64) bool {
}
// Returns true if the error was sent, or false if watcher is closed.
func (w *Watcher) sendError(err error) bool {
func (w *readDirChangesW) sendError(err error) bool {
if err == nil {
return true
}
select {
case w.Errors <- err:
return true
case <-w.quit:
return false
}
return false
}
// Close removes all watches and closes the events channel.
func (w *Watcher) Close() error {
w.mu.Lock()
if w.isClosed {
w.mu.Unlock()
func (w *readDirChangesW) Close() error {
if w.isClosed() {
return nil
}
w.isClosed = true
w.mu.Lock()
w.closed = true
w.mu.Unlock()
// Send "quit" message to the reader goroutine
@ -186,49 +108,31 @@ func (w *Watcher) Close() error {
return <-ch
}
// Add starts monitoring the path for changes.
//
// A path can only be watched once; attempting to watch it more than once will
// return an error. Paths that do not yet exist on the filesystem cannot be
// added. A watch will be automatically removed if the path is deleted.
//
// A path will remain watched if it gets renamed to somewhere else on the same
// filesystem, but the monitor will get removed if the path gets deleted and
// re-created, or if it's moved to a different filesystem.
//
// Notifications on network filesystems (NFS, SMB, FUSE, etc.) or special
// filesystems (/proc, /sys, etc.) generally don't work.
//
// # Watching directories
//
// All files in a directory are monitored, including new files that are created
// after the watcher is started. Subdirectories are not watched (i.e. it's
// non-recursive).
//
// # Watching files
//
// Watching individual files (rather than directories) is generally not
// recommended as many tools update files atomically. Instead of "just" writing
// to the file a temporary file will be written to first, and if successful the
// temporary file is moved to to destination removing the original, or some
// variant thereof. The watcher on the original file is now lost, as it no
// longer exists.
//
// Instead, watch the parent directory and use Event.Name to filter out files
// you're not interested in. There is an example of this in [cmd/fsnotify/file.go].
func (w *Watcher) Add(name string) error {
w.mu.Lock()
if w.isClosed {
w.mu.Unlock()
return errors.New("watcher already closed")
func (w *readDirChangesW) Add(name string) error { return w.AddWith(name) }
func (w *readDirChangesW) AddWith(name string, opts ...addOpt) error {
if w.isClosed() {
return ErrClosed
}
if debug {
fmt.Fprintf(os.Stderr, "FSNOTIFY_DEBUG: %s AddWith(%q)\n",
time.Now().Format("15:04:05.000000000"), filepath.ToSlash(name))
}
with := getOptions(opts...)
if !w.xSupports(with.op) {
return fmt.Errorf("%w: %s", xErrUnsupported, with.op)
}
if with.bufsize < 4096 {
return fmt.Errorf("fsnotify.WithBufferSize: buffer size cannot be smaller than 4096 bytes")
}
w.mu.Unlock()
in := &input{
op: opAddWatch,
path: filepath.Clean(name),
flags: sysFSALLEVENTS,
reply: make(chan error),
op: opAddWatch,
path: filepath.Clean(name),
flags: sysFSALLEVENTS,
reply: make(chan error),
bufsize: with.bufsize,
}
w.input <- in
if err := w.wakeupReader(); err != nil {
@ -237,13 +141,15 @@ func (w *Watcher) Add(name string) error {
return <-in.reply
}
// Remove stops monitoring the path for changes.
//
// Directories are always removed non-recursively. For example, if you added
// /tmp/dir and /tmp/dir/subdir then you will need to remove both.
//
// Removing a path that has not yet been added returns [ErrNonExistentWatch].
func (w *Watcher) Remove(name string) error {
func (w *readDirChangesW) Remove(name string) error {
if w.isClosed() {
return nil
}
if debug {
fmt.Fprintf(os.Stderr, "FSNOTIFY_DEBUG: %s Remove(%q)\n",
time.Now().Format("15:04:05.000000000"), filepath.ToSlash(name))
}
in := &input{
op: opRemoveWatch,
path: filepath.Clean(name),
@ -256,15 +162,24 @@ func (w *Watcher) Remove(name string) error {
return <-in.reply
}
// WatchList returns all paths added with [Add] (and are not yet removed).
func (w *Watcher) WatchList() []string {
func (w *readDirChangesW) WatchList() []string {
if w.isClosed() {
return nil
}
w.mu.Lock()
defer w.mu.Unlock()
entries := make([]string, 0, len(w.watches))
for _, entry := range w.watches {
for _, watchEntry := range entry {
entries = append(entries, watchEntry.path)
for name := range watchEntry.names {
entries = append(entries, filepath.Join(watchEntry.path, name))
}
// the directory itself is being watched
if watchEntry.mask != 0 {
entries = append(entries, watchEntry.path)
}
}
}
@ -279,7 +194,6 @@ func (w *Watcher) WatchList() []string {
// This should all be removed at some point, and just use windows.FILE_NOTIFY_*
const (
sysFSALLEVENTS = 0xfff
sysFSATTRIB = 0x4
sysFSCREATE = 0x100
sysFSDELETE = 0x200
sysFSDELETESELF = 0x400
@ -291,7 +205,7 @@ const (
sysFSIGNORED = 0x8000
)
func (w *Watcher) newEvent(name string, mask uint32) Event {
func (w *readDirChangesW) newEvent(name string, mask uint32) Event {
e := Event{Name: name}
if mask&sysFSCREATE == sysFSCREATE || mask&sysFSMOVEDTO == sysFSMOVEDTO {
e.Op |= Create
@ -305,9 +219,6 @@ func (w *Watcher) newEvent(name string, mask uint32) Event {
if mask&sysFSMOVE == sysFSMOVE || mask&sysFSMOVESELF == sysFSMOVESELF || mask&sysFSMOVEDFROM == sysFSMOVEDFROM {
e.Op |= Rename
}
if mask&sysFSATTRIB == sysFSATTRIB {
e.Op |= Chmod
}
return e
}
@ -321,10 +232,11 @@ const (
)
type input struct {
op int
path string
flags uint32
reply chan error
op int
path string
flags uint32
bufsize int
reply chan error
}
type inode struct {
@ -334,13 +246,14 @@ type inode struct {
}
type watch struct {
ov windows.Overlapped
ino *inode // i-number
path string // Directory path
mask uint64 // Directory itself is being watched with these notify flags
names map[string]uint64 // Map of names being watched and their notify flags
rename string // Remembers the old name while renaming a file
buf [65536]byte // 64K buffer
ov windows.Overlapped
ino *inode // i-number
recurse bool // Recursive watch?
path string // Directory path
mask uint64 // Directory itself is being watched with these notify flags
names map[string]uint64 // Map of names being watched and their notify flags
rename string // Remembers the old name while renaming a file
buf []byte // buffer, allocated later
}
type (
@ -348,7 +261,7 @@ type (
watchMap map[uint32]indexMap
)
func (w *Watcher) wakeupReader() error {
func (w *readDirChangesW) wakeupReader() error {
err := windows.PostQueuedCompletionStatus(w.port, 0, 0, nil)
if err != nil {
return os.NewSyscallError("PostQueuedCompletionStatus", err)
@ -356,7 +269,7 @@ func (w *Watcher) wakeupReader() error {
return nil
}
func (w *Watcher) getDir(pathname string) (dir string, err error) {
func (w *readDirChangesW) getDir(pathname string) (dir string, err error) {
attr, err := windows.GetFileAttributes(windows.StringToUTF16Ptr(pathname))
if err != nil {
return "", os.NewSyscallError("GetFileAttributes", err)
@ -370,7 +283,7 @@ func (w *Watcher) getDir(pathname string) (dir string, err error) {
return
}
func (w *Watcher) getIno(path string) (ino *inode, err error) {
func (w *readDirChangesW) getIno(path string) (ino *inode, err error) {
h, err := windows.CreateFile(windows.StringToUTF16Ptr(path),
windows.FILE_LIST_DIRECTORY,
windows.FILE_SHARE_READ|windows.FILE_SHARE_WRITE|windows.FILE_SHARE_DELETE,
@ -413,7 +326,9 @@ func (m watchMap) set(ino *inode, watch *watch) {
}
// Must run within the I/O thread.
func (w *Watcher) addWatch(pathname string, flags uint64) error {
func (w *readDirChangesW) addWatch(pathname string, flags uint64, bufsize int) error {
pathname, recurse := recursivePath(pathname)
dir, err := w.getDir(pathname)
if err != nil {
return err
@ -433,9 +348,11 @@ func (w *Watcher) addWatch(pathname string, flags uint64) error {
return os.NewSyscallError("CreateIoCompletionPort", err)
}
watchEntry = &watch{
ino: ino,
path: dir,
names: make(map[string]uint64),
ino: ino,
path: dir,
names: make(map[string]uint64),
recurse: recurse,
buf: make([]byte, bufsize),
}
w.mu.Lock()
w.watches.set(ino, watchEntry)
@ -464,7 +381,9 @@ func (w *Watcher) addWatch(pathname string, flags uint64) error {
}
// Must run within the I/O thread.
func (w *Watcher) remWatch(pathname string) error {
func (w *readDirChangesW) remWatch(pathname string) error {
pathname, recurse := recursivePath(pathname)
dir, err := w.getDir(pathname)
if err != nil {
return err
@ -478,6 +397,10 @@ func (w *Watcher) remWatch(pathname string) error {
watch := w.watches.get(ino)
w.mu.Unlock()
if recurse && !watch.recurse {
return fmt.Errorf("can't use \\... with non-recursive watch %q", pathname)
}
err = windows.CloseHandle(ino.handle)
if err != nil {
w.sendError(os.NewSyscallError("CloseHandle", err))
@ -486,11 +409,11 @@ func (w *Watcher) remWatch(pathname string) error {
return fmt.Errorf("%w: %s", ErrNonExistentWatch, pathname)
}
if pathname == dir {
w.sendEvent(watch.path, watch.mask&sysFSIGNORED)
w.sendEvent(watch.path, "", watch.mask&sysFSIGNORED)
watch.mask = 0
} else {
name := filepath.Base(pathname)
w.sendEvent(filepath.Join(watch.path, name), watch.names[name]&sysFSIGNORED)
w.sendEvent(filepath.Join(watch.path, name), "", watch.names[name]&sysFSIGNORED)
delete(watch.names, name)
}
@ -498,23 +421,23 @@ func (w *Watcher) remWatch(pathname string) error {
}
// Must run within the I/O thread.
func (w *Watcher) deleteWatch(watch *watch) {
func (w *readDirChangesW) deleteWatch(watch *watch) {
for name, mask := range watch.names {
if mask&provisional == 0 {
w.sendEvent(filepath.Join(watch.path, name), mask&sysFSIGNORED)
w.sendEvent(filepath.Join(watch.path, name), "", mask&sysFSIGNORED)
}
delete(watch.names, name)
}
if watch.mask != 0 {
if watch.mask&provisional == 0 {
w.sendEvent(watch.path, watch.mask&sysFSIGNORED)
w.sendEvent(watch.path, "", watch.mask&sysFSIGNORED)
}
watch.mask = 0
}
}
// Must run within the I/O thread.
func (w *Watcher) startRead(watch *watch) error {
func (w *readDirChangesW) startRead(watch *watch) error {
err := windows.CancelIo(watch.ino.handle)
if err != nil {
w.sendError(os.NewSyscallError("CancelIo", err))
@ -535,13 +458,16 @@ func (w *Watcher) startRead(watch *watch) error {
return nil
}
rdErr := windows.ReadDirectoryChanges(watch.ino.handle, &watch.buf[0],
uint32(unsafe.Sizeof(watch.buf)), false, mask, nil, &watch.ov, 0)
// We need to pass the array, rather than the slice.
hdr := (*reflect.SliceHeader)(unsafe.Pointer(&watch.buf))
rdErr := windows.ReadDirectoryChanges(watch.ino.handle,
(*byte)(unsafe.Pointer(hdr.Data)), uint32(hdr.Len),
watch.recurse, mask, nil, &watch.ov, 0)
if rdErr != nil {
err := os.NewSyscallError("ReadDirectoryChanges", rdErr)
if rdErr == windows.ERROR_ACCESS_DENIED && watch.mask&provisional == 0 {
// Watched directory was probably removed
w.sendEvent(watch.path, watch.mask&sysFSDELETESELF)
w.sendEvent(watch.path, "", watch.mask&sysFSDELETESELF)
err = nil
}
w.deleteWatch(watch)
@ -554,7 +480,7 @@ func (w *Watcher) startRead(watch *watch) error {
// readEvents reads from the I/O completion port, converts the
// received events into Event objects and sends them via the Events channel.
// Entry point to the I/O thread.
func (w *Watcher) readEvents() {
func (w *readDirChangesW) readEvents() {
var (
n uint32
key uintptr
@ -563,9 +489,8 @@ func (w *Watcher) readEvents() {
runtime.LockOSThread()
for {
// This error is handled after the watch == nil check below.
qErr := windows.GetQueuedCompletionStatus(w.port, &n, &key, &ov, windows.INFINITE)
// This error is handled after the watch == nil check below. NOTE: this
// seems odd, note sure if it's correct.
watch := (*watch)(unsafe.Pointer(ov))
if watch == nil {
@ -595,7 +520,7 @@ func (w *Watcher) readEvents() {
case in := <-w.input:
switch in.op {
case opAddWatch:
in.reply <- w.addWatch(in.path, uint64(in.flags))
in.reply <- w.addWatch(in.path, uint64(in.flags), in.bufsize)
case opRemoveWatch:
in.reply <- w.remWatch(in.path)
}
@ -605,6 +530,8 @@ func (w *Watcher) readEvents() {
}
switch qErr {
case nil:
// No error
case windows.ERROR_MORE_DATA:
if watch == nil {
w.sendError(errors.New("ERROR_MORE_DATA has unexpectedly null lpOverlapped buffer"))
@ -616,7 +543,7 @@ func (w *Watcher) readEvents() {
}
case windows.ERROR_ACCESS_DENIED:
// Watched directory was probably removed
w.sendEvent(watch.path, watch.mask&sysFSDELETESELF)
w.sendEvent(watch.path, "", watch.mask&sysFSDELETESELF)
w.deleteWatch(watch)
w.startRead(watch)
continue
@ -626,13 +553,12 @@ func (w *Watcher) readEvents() {
default:
w.sendError(os.NewSyscallError("GetQueuedCompletionPort", qErr))
continue
case nil:
}
var offset uint32
for {
if n == 0 {
w.sendError(errors.New("short read in readEvents()"))
w.sendError(ErrEventOverflow)
break
}
@ -650,6 +576,10 @@ func (w *Watcher) readEvents() {
name := windows.UTF16ToString(buf)
fullname := filepath.Join(watch.path, name)
if debug {
internal.Debug(fullname, raw.Action)
}
var mask uint64
switch raw.Action {
case windows.FILE_ACTION_REMOVED:
@ -678,21 +608,22 @@ func (w *Watcher) readEvents() {
}
}
sendNameEvent := func() {
w.sendEvent(fullname, watch.names[name]&mask)
}
if raw.Action != windows.FILE_ACTION_RENAMED_NEW_NAME {
sendNameEvent()
w.sendEvent(fullname, "", watch.names[name]&mask)
}
if raw.Action == windows.FILE_ACTION_REMOVED {
w.sendEvent(fullname, watch.names[name]&sysFSIGNORED)
w.sendEvent(fullname, "", watch.names[name]&sysFSIGNORED)
delete(watch.names, name)
}
w.sendEvent(fullname, watch.mask&w.toFSnotifyFlags(raw.Action))
if watch.rename != "" && raw.Action == windows.FILE_ACTION_RENAMED_NEW_NAME {
w.sendEvent(fullname, filepath.Join(watch.path, watch.rename), watch.mask&w.toFSnotifyFlags(raw.Action))
} else {
w.sendEvent(fullname, "", watch.mask&w.toFSnotifyFlags(raw.Action))
}
if raw.Action == windows.FILE_ACTION_RENAMED_NEW_NAME {
fullname = filepath.Join(watch.path, watch.rename)
sendNameEvent()
w.sendEvent(filepath.Join(watch.path, watch.rename), "", watch.names[name]&mask)
}
// Move to the next event in the buffer
@ -703,8 +634,8 @@ func (w *Watcher) readEvents() {
// Error!
if offset >= n {
w.sendError(errors.New(
"Windows system assumed buffer larger than it is, events have likely been missed."))
//lint:ignore ST1005 Windows should be capitalized
w.sendError(errors.New("Windows system assumed buffer larger than it is, events have likely been missed"))
break
}
}
@ -715,21 +646,18 @@ func (w *Watcher) readEvents() {
}
}
func (w *Watcher) toWindowsFlags(mask uint64) uint32 {
func (w *readDirChangesW) toWindowsFlags(mask uint64) uint32 {
var m uint32
if mask&sysFSMODIFY != 0 {
m |= windows.FILE_NOTIFY_CHANGE_LAST_WRITE
}
if mask&sysFSATTRIB != 0 {
m |= windows.FILE_NOTIFY_CHANGE_ATTRIBUTES
}
if mask&(sysFSMOVE|sysFSCREATE|sysFSDELETE) != 0 {
m |= windows.FILE_NOTIFY_CHANGE_FILE_NAME | windows.FILE_NOTIFY_CHANGE_DIR_NAME
}
return m
}
func (w *Watcher) toFSnotifyFlags(action uint32) uint64 {
func (w *readDirChangesW) toFSnotifyFlags(action uint32) uint64 {
switch action {
case windows.FILE_ACTION_ADDED:
return sysFSCREATE
@ -744,3 +672,11 @@ func (w *Watcher) toFSnotifyFlags(action uint32) uint64 {
}
return 0
}
func (w *readDirChangesW) xSupports(op Op) bool {
if op.Has(xUnportableOpen) || op.Has(xUnportableRead) ||
op.Has(xUnportableCloseWrite) || op.Has(xUnportableCloseRead) {
return false
}
return true
}

View file

@ -1,16 +1,148 @@
//go:build !plan9
// +build !plan9
// Package fsnotify provides a cross-platform interface for file system
// notifications.
//
// Currently supported systems:
//
// - Linux via inotify
// - BSD, macOS via kqueue
// - Windows via ReadDirectoryChangesW
// - illumos via FEN
//
// # FSNOTIFY_DEBUG
//
// Set the FSNOTIFY_DEBUG environment variable to "1" to print debug messages to
// stderr. This can be useful to track down some problems, especially in cases
// where fsnotify is used as an indirect dependency.
//
// Every event will be printed as soon as there's something useful to print,
// with as little processing from fsnotify.
//
// Example output:
//
// FSNOTIFY_DEBUG: 11:34:23.633087586 256:IN_CREATE → "/tmp/file-1"
// FSNOTIFY_DEBUG: 11:34:23.633202319 4:IN_ATTRIB → "/tmp/file-1"
// FSNOTIFY_DEBUG: 11:34:28.989728764 512:IN_DELETE → "/tmp/file-1"
package fsnotify
import (
"errors"
"fmt"
"os"
"path/filepath"
"strings"
)
// Watcher watches a set of paths, delivering events on a channel.
//
// A watcher should not be copied (e.g. pass it by pointer, rather than by
// value).
//
// # Linux notes
//
// When a file is removed a Remove event won't be emitted until all file
// descriptors are closed, and deletes will always emit a Chmod. For example:
//
// fp := os.Open("file")
// os.Remove("file") // Triggers Chmod
// fp.Close() // Triggers Remove
//
// This is the event that inotify sends, so not much can be changed about this.
//
// The fs.inotify.max_user_watches sysctl variable specifies the upper limit
// for the number of watches per user, and fs.inotify.max_user_instances
// specifies the maximum number of inotify instances per user. Every Watcher you
// create is an "instance", and every path you add is a "watch".
//
// These are also exposed in /proc as /proc/sys/fs/inotify/max_user_watches and
// /proc/sys/fs/inotify/max_user_instances
//
// To increase them you can use sysctl or write the value to the /proc file:
//
// # Default values on Linux 5.18
// sysctl fs.inotify.max_user_watches=124983
// sysctl fs.inotify.max_user_instances=128
//
// To make the changes persist on reboot edit /etc/sysctl.conf or
// /usr/lib/sysctl.d/50-default.conf (details differ per Linux distro; check
// your distro's documentation):
//
// fs.inotify.max_user_watches=124983
// fs.inotify.max_user_instances=128
//
// Reaching the limit will result in a "no space left on device" or "too many open
// files" error.
//
// # kqueue notes (macOS, BSD)
//
// kqueue requires opening a file descriptor for every file that's being watched;
// so if you're watching a directory with five files then that's six file
// descriptors. You will run in to your system's "max open files" limit faster on
// these platforms.
//
// The sysctl variables kern.maxfiles and kern.maxfilesperproc can be used to
// control the maximum number of open files, as well as /etc/login.conf on BSD
// systems.
//
// # Windows notes
//
// Paths can be added as "C:\\path\\to\\dir", but forward slashes
// ("C:/path/to/dir") will also work.
//
// When a watched directory is removed it will always send an event for the
// directory itself, but may not send events for all files in that directory.
// Sometimes it will send events for all files, sometimes it will send no
// events, and often only for some files.
//
// The default ReadDirectoryChangesW() buffer size is 64K, which is the largest
// value that is guaranteed to work with SMB filesystems. If you have many
// events in quick succession this may not be enough, and you will have to use
// [WithBufferSize] to increase the value.
type Watcher struct {
b backend
// Events sends the filesystem change events.
//
// fsnotify can send the following events; a "path" here can refer to a
// file, directory, symbolic link, or special file like a FIFO.
//
// fsnotify.Create A new path was created; this may be followed by one
// or more Write events if data also gets written to a
// file.
//
// fsnotify.Remove A path was removed.
//
// fsnotify.Rename A path was renamed. A rename is always sent with the
// old path as Event.Name, and a Create event will be
// sent with the new name. Renames are only sent for
// paths that are currently watched; e.g. moving an
// unmonitored file into a monitored directory will
// show up as just a Create. Similarly, renaming a file
// to outside a monitored directory will show up as
// only a Rename.
//
// fsnotify.Write A file or named pipe was written to. A Truncate will
// also trigger a Write. A single "write action"
// initiated by the user may show up as one or multiple
// writes, depending on when the system syncs things to
// disk. For example when compiling a large Go program
// you may get hundreds of Write events, and you may
// want to wait until you've stopped receiving them
// (see the dedup example in cmd/fsnotify).
//
// Some systems may send Write event for directories
// when the directory content changes.
//
// fsnotify.Chmod Attributes were changed. On Linux this is also sent
// when a file is removed (or more accurately, when a
// link to an inode is removed). On kqueue it's sent
// when a file is truncated. On Windows it's never
// sent.
Events chan Event
// Errors sends any errors.
Errors chan error
}
// Event represents a file system notification.
type Event struct {
// Path to the file or directory.
@ -25,6 +157,16 @@ type Event struct {
// This is a bitmask and some systems may send multiple operations at once.
// Use the Event.Has() method instead of comparing with ==.
Op Op
// Create events will have this set to the old path if it's a rename. This
// only works when both the source and destination are watched. It's not
// reliable when watching individual files, only directories.
//
// For example "mv /tmp/file /tmp/rename" will emit:
//
// Event{Op: Rename, Name: "/tmp/file"}
// Event{Op: Create, Name: "/tmp/rename", RenamedFrom: "/tmp/file"}
renamedFrom string
}
// Op describes a set of file operations.
@ -33,34 +175,204 @@ type Op uint32
// The operations fsnotify can trigger; see the documentation on [Watcher] for a
// full description, and check them with [Event.Has].
const (
// A new pathname was created.
Create Op = 1 << iota
// The pathname was written to; this does *not* mean the write has finished,
// and a write can be followed by more writes.
Write
// The path was removed; any watches on it will be removed. Some "remove"
// operations may trigger a Rename if the file is actually moved (for
// example "remove to trash" is often a rename).
Remove
// The path was renamed to something else; any watches on it will be
// removed.
Rename
// File attributes were changed.
//
// It's generally not recommended to take action on this event, as it may
// get triggered very frequently by some software. For example, Spotlight
// indexing on macOS, anti-virus software, backup software, etc.
Chmod
// File descriptor was opened.
//
// Only works on Linux and FreeBSD.
xUnportableOpen
// File was read from.
//
// Only works on Linux and FreeBSD.
xUnportableRead
// File opened for writing was closed.
//
// Only works on Linux and FreeBSD.
//
// The advantage of using this over Write is that it's more reliable than
// waiting for Write events to stop. It's also faster (if you're not
// listening to Write events): copying a file of a few GB can easily
// generate tens of thousands of Write events in a short span of time.
xUnportableCloseWrite
// File opened for reading was closed.
//
// Only works on Linux and FreeBSD.
xUnportableCloseRead
)
// Common errors that can be reported by a watcher
var (
ErrNonExistentWatch = errors.New("can't remove non-existent watcher")
ErrEventOverflow = errors.New("fsnotify queue overflow")
// ErrNonExistentWatch is used when Remove() is called on a path that's not
// added.
ErrNonExistentWatch = errors.New("fsnotify: can't remove non-existent watch")
// ErrClosed is used when trying to operate on a closed Watcher.
ErrClosed = errors.New("fsnotify: watcher already closed")
// ErrEventOverflow is reported from the Errors channel when there are too
// many events:
//
// - inotify: inotify returns IN_Q_OVERFLOW because there are too
// many queued events (the fs.inotify.max_queued_events
// sysctl can be used to increase this).
// - windows: The buffer size is too small; WithBufferSize() can be used to increase it.
// - kqueue, fen: Not used.
ErrEventOverflow = errors.New("fsnotify: queue or buffer overflow")
// ErrUnsupported is returned by AddWith() when WithOps() specified an
// Unportable event that's not supported on this platform.
xErrUnsupported = errors.New("fsnotify: not supported with this backend")
)
func (op Op) String() string {
// NewWatcher creates a new Watcher.
func NewWatcher() (*Watcher, error) {
ev, errs := make(chan Event), make(chan error)
b, err := newBackend(ev, errs)
if err != nil {
return nil, err
}
return &Watcher{b: b, Events: ev, Errors: errs}, nil
}
// NewBufferedWatcher creates a new Watcher with a buffered Watcher.Events
// channel.
//
// The main use case for this is situations with a very large number of events
// where the kernel buffer size can't be increased (e.g. due to lack of
// permissions). An unbuffered Watcher will perform better for almost all use
// cases, and whenever possible you will be better off increasing the kernel
// buffers instead of adding a large userspace buffer.
func NewBufferedWatcher(sz uint) (*Watcher, error) {
ev, errs := make(chan Event), make(chan error)
b, err := newBufferedBackend(sz, ev, errs)
if err != nil {
return nil, err
}
return &Watcher{b: b, Events: ev, Errors: errs}, nil
}
// Add starts monitoring the path for changes.
//
// A path can only be watched once; watching it more than once is a no-op and will
// not return an error. Paths that do not yet exist on the filesystem cannot be
// watched.
//
// A watch will be automatically removed if the watched path is deleted or
// renamed. The exception is the Windows backend, which doesn't remove the
// watcher on renames.
//
// Notifications on network filesystems (NFS, SMB, FUSE, etc.) or special
// filesystems (/proc, /sys, etc.) generally don't work.
//
// Returns [ErrClosed] if [Watcher.Close] was called.
//
// See [Watcher.AddWith] for a version that allows adding options.
//
// # Watching directories
//
// All files in a directory are monitored, including new files that are created
// after the watcher is started. Subdirectories are not watched (i.e. it's
// non-recursive).
//
// # Watching files
//
// Watching individual files (rather than directories) is generally not
// recommended as many programs (especially editors) update files atomically: it
// will write to a temporary file which is then moved to destination,
// overwriting the original (or some variant thereof). The watcher on the
// original file is now lost, as that no longer exists.
//
// The upshot of this is that a power failure or crash won't leave a
// half-written file.
//
// Watch the parent directory and use Event.Name to filter out files you're not
// interested in. There is an example of this in cmd/fsnotify/file.go.
func (w *Watcher) Add(path string) error { return w.b.Add(path) }
// AddWith is like [Watcher.Add], but allows adding options. When using Add()
// the defaults described below are used.
//
// Possible options are:
//
// - [WithBufferSize] sets the buffer size for the Windows backend; no-op on
// other platforms. The default is 64K (65536 bytes).
func (w *Watcher) AddWith(path string, opts ...addOpt) error { return w.b.AddWith(path, opts...) }
// Remove stops monitoring the path for changes.
//
// Directories are always removed non-recursively. For example, if you added
// /tmp/dir and /tmp/dir/subdir then you will need to remove both.
//
// Removing a path that has not yet been added returns [ErrNonExistentWatch].
//
// Returns nil if [Watcher.Close] was called.
func (w *Watcher) Remove(path string) error { return w.b.Remove(path) }
// Close removes all watches and closes the Events channel.
func (w *Watcher) Close() error { return w.b.Close() }
// WatchList returns all paths explicitly added with [Watcher.Add] (and are not
// yet removed).
//
// Returns nil if [Watcher.Close] was called.
func (w *Watcher) WatchList() []string { return w.b.WatchList() }
// Supports reports if all the listed operations are supported by this platform.
//
// Create, Write, Remove, Rename, and Chmod are always supported. It can only
// return false for an Op starting with Unportable.
func (w *Watcher) xSupports(op Op) bool { return w.b.xSupports(op) }
func (o Op) String() string {
var b strings.Builder
if op.Has(Create) {
if o.Has(Create) {
b.WriteString("|CREATE")
}
if op.Has(Remove) {
if o.Has(Remove) {
b.WriteString("|REMOVE")
}
if op.Has(Write) {
if o.Has(Write) {
b.WriteString("|WRITE")
}
if op.Has(Rename) {
if o.Has(xUnportableOpen) {
b.WriteString("|OPEN")
}
if o.Has(xUnportableRead) {
b.WriteString("|READ")
}
if o.Has(xUnportableCloseWrite) {
b.WriteString("|CLOSE_WRITE")
}
if o.Has(xUnportableCloseRead) {
b.WriteString("|CLOSE_READ")
}
if o.Has(Rename) {
b.WriteString("|RENAME")
}
if op.Has(Chmod) {
if o.Has(Chmod) {
b.WriteString("|CHMOD")
}
if b.Len() == 0 {
@ -70,12 +382,113 @@ func (op Op) String() string {
}
// Has reports if this operation has the given operation.
func (o Op) Has(h Op) bool { return o&h == h }
func (o Op) Has(h Op) bool { return o&h != 0 }
// Has reports if this event has the given operation.
func (e Event) Has(op Op) bool { return e.Op.Has(op) }
// String returns a string representation of the event with their path.
func (e Event) String() string {
if e.renamedFrom != "" {
return fmt.Sprintf("%-13s %q ← %q", e.Op.String(), e.Name, e.renamedFrom)
}
return fmt.Sprintf("%-13s %q", e.Op.String(), e.Name)
}
type (
backend interface {
Add(string) error
AddWith(string, ...addOpt) error
Remove(string) error
WatchList() []string
Close() error
xSupports(Op) bool
}
addOpt func(opt *withOpts)
withOpts struct {
bufsize int
op Op
noFollow bool
sendCreate bool
}
)
var debug = func() bool {
// Check for exactly "1" (rather than mere existence) so we can add
// options/flags in the future. I don't know if we ever want that, but it's
// nice to leave the option open.
return os.Getenv("FSNOTIFY_DEBUG") == "1"
}()
var defaultOpts = withOpts{
bufsize: 65536, // 64K
op: Create | Write | Remove | Rename | Chmod,
}
func getOptions(opts ...addOpt) withOpts {
with := defaultOpts
for _, o := range opts {
if o != nil {
o(&with)
}
}
return with
}
// WithBufferSize sets the [ReadDirectoryChangesW] buffer size.
//
// This only has effect on Windows systems, and is a no-op for other backends.
//
// The default value is 64K (65536 bytes) which is the highest value that works
// on all filesystems and should be enough for most applications, but if you
// have a large burst of events it may not be enough. You can increase it if
// you're hitting "queue or buffer overflow" errors ([ErrEventOverflow]).
//
// [ReadDirectoryChangesW]: https://learn.microsoft.com/en-gb/windows/win32/api/winbase/nf-winbase-readdirectorychangesw
func WithBufferSize(bytes int) addOpt {
return func(opt *withOpts) { opt.bufsize = bytes }
}
// WithOps sets which operations to listen for. The default is [Create],
// [Write], [Remove], [Rename], and [Chmod].
//
// Excluding operations you're not interested in can save quite a bit of CPU
// time; in some use cases there may be hundreds of thousands of useless Write
// or Chmod operations per second.
//
// This can also be used to add unportable operations not supported by all
// platforms; unportable operations all start with "Unportable":
// [UnportableOpen], [UnportableRead], [UnportableCloseWrite], and
// [UnportableCloseRead].
//
// AddWith returns an error when using an unportable operation that's not
// supported. Use [Watcher.Support] to check for support.
func withOps(op Op) addOpt {
return func(opt *withOpts) { opt.op = op }
}
// WithNoFollow disables following symlinks, so the symlinks themselves are
// watched.
func withNoFollow() addOpt {
return func(opt *withOpts) { opt.noFollow = true }
}
// "Internal" option for recursive watches on inotify.
func withCreate() addOpt {
return func(opt *withOpts) { opt.sendCreate = true }
}
var enableRecurse = false
// Check if this path is recursive (ends with "/..." or "\..."), and return the
// path with the /... stripped.
func recursivePath(path string) (string, bool) {
path = filepath.Clean(path)
if !enableRecurse { // Only enabled in tests for now.
return path, false
}
if filepath.Base(path) == "..." {
return filepath.Dir(path), true
}
return path, false
}

39
vendor/github.com/fsnotify/fsnotify/internal/darwin.go generated vendored Normal file
View file

@ -0,0 +1,39 @@
//go:build darwin
package internal
import (
"syscall"
"golang.org/x/sys/unix"
)
var (
SyscallEACCES = syscall.EACCES
UnixEACCES = unix.EACCES
)
var maxfiles uint64
// Go 1.19 will do this automatically: https://go-review.googlesource.com/c/go/+/393354/
func SetRlimit() {
var l syscall.Rlimit
err := syscall.Getrlimit(syscall.RLIMIT_NOFILE, &l)
if err == nil && l.Cur != l.Max {
l.Cur = l.Max
syscall.Setrlimit(syscall.RLIMIT_NOFILE, &l)
}
maxfiles = l.Cur
if n, err := syscall.SysctlUint32("kern.maxfiles"); err == nil && uint64(n) < maxfiles {
maxfiles = uint64(n)
}
if n, err := syscall.SysctlUint32("kern.maxfilesperproc"); err == nil && uint64(n) < maxfiles {
maxfiles = uint64(n)
}
}
func Maxfiles() uint64 { return maxfiles }
func Mkfifo(path string, mode uint32) error { return unix.Mkfifo(path, mode) }
func Mknod(path string, mode uint32, dev int) error { return unix.Mknod(path, mode, dev) }

View file

@ -0,0 +1,57 @@
package internal
import "golang.org/x/sys/unix"
var names = []struct {
n string
m uint32
}{
{"NOTE_ABSOLUTE", unix.NOTE_ABSOLUTE},
{"NOTE_ATTRIB", unix.NOTE_ATTRIB},
{"NOTE_BACKGROUND", unix.NOTE_BACKGROUND},
{"NOTE_CHILD", unix.NOTE_CHILD},
{"NOTE_CRITICAL", unix.NOTE_CRITICAL},
{"NOTE_DELETE", unix.NOTE_DELETE},
{"NOTE_EXEC", unix.NOTE_EXEC},
{"NOTE_EXIT", unix.NOTE_EXIT},
{"NOTE_EXITSTATUS", unix.NOTE_EXITSTATUS},
{"NOTE_EXIT_CSERROR", unix.NOTE_EXIT_CSERROR},
{"NOTE_EXIT_DECRYPTFAIL", unix.NOTE_EXIT_DECRYPTFAIL},
{"NOTE_EXIT_DETAIL", unix.NOTE_EXIT_DETAIL},
{"NOTE_EXIT_DETAIL_MASK", unix.NOTE_EXIT_DETAIL_MASK},
{"NOTE_EXIT_MEMORY", unix.NOTE_EXIT_MEMORY},
{"NOTE_EXIT_REPARENTED", unix.NOTE_EXIT_REPARENTED},
{"NOTE_EXTEND", unix.NOTE_EXTEND},
{"NOTE_FFAND", unix.NOTE_FFAND},
{"NOTE_FFCOPY", unix.NOTE_FFCOPY},
{"NOTE_FFCTRLMASK", unix.NOTE_FFCTRLMASK},
{"NOTE_FFLAGSMASK", unix.NOTE_FFLAGSMASK},
{"NOTE_FFNOP", unix.NOTE_FFNOP},
{"NOTE_FFOR", unix.NOTE_FFOR},
{"NOTE_FORK", unix.NOTE_FORK},
{"NOTE_FUNLOCK", unix.NOTE_FUNLOCK},
{"NOTE_LEEWAY", unix.NOTE_LEEWAY},
{"NOTE_LINK", unix.NOTE_LINK},
{"NOTE_LOWAT", unix.NOTE_LOWAT},
{"NOTE_MACHTIME", unix.NOTE_MACHTIME},
{"NOTE_MACH_CONTINUOUS_TIME", unix.NOTE_MACH_CONTINUOUS_TIME},
{"NOTE_NONE", unix.NOTE_NONE},
{"NOTE_NSECONDS", unix.NOTE_NSECONDS},
{"NOTE_OOB", unix.NOTE_OOB},
//{"NOTE_PCTRLMASK", unix.NOTE_PCTRLMASK}, -0x100000 (?!)
{"NOTE_PDATAMASK", unix.NOTE_PDATAMASK},
{"NOTE_REAP", unix.NOTE_REAP},
{"NOTE_RENAME", unix.NOTE_RENAME},
{"NOTE_REVOKE", unix.NOTE_REVOKE},
{"NOTE_SECONDS", unix.NOTE_SECONDS},
{"NOTE_SIGNAL", unix.NOTE_SIGNAL},
{"NOTE_TRACK", unix.NOTE_TRACK},
{"NOTE_TRACKERR", unix.NOTE_TRACKERR},
{"NOTE_TRIGGER", unix.NOTE_TRIGGER},
{"NOTE_USECONDS", unix.NOTE_USECONDS},
{"NOTE_VM_ERROR", unix.NOTE_VM_ERROR},
{"NOTE_VM_PRESSURE", unix.NOTE_VM_PRESSURE},
{"NOTE_VM_PRESSURE_SUDDEN_TERMINATE", unix.NOTE_VM_PRESSURE_SUDDEN_TERMINATE},
{"NOTE_VM_PRESSURE_TERMINATE", unix.NOTE_VM_PRESSURE_TERMINATE},
{"NOTE_WRITE", unix.NOTE_WRITE},
}

View file

@ -0,0 +1,33 @@
package internal
import "golang.org/x/sys/unix"
var names = []struct {
n string
m uint32
}{
{"NOTE_ATTRIB", unix.NOTE_ATTRIB},
{"NOTE_CHILD", unix.NOTE_CHILD},
{"NOTE_DELETE", unix.NOTE_DELETE},
{"NOTE_EXEC", unix.NOTE_EXEC},
{"NOTE_EXIT", unix.NOTE_EXIT},
{"NOTE_EXTEND", unix.NOTE_EXTEND},
{"NOTE_FFAND", unix.NOTE_FFAND},
{"NOTE_FFCOPY", unix.NOTE_FFCOPY},
{"NOTE_FFCTRLMASK", unix.NOTE_FFCTRLMASK},
{"NOTE_FFLAGSMASK", unix.NOTE_FFLAGSMASK},
{"NOTE_FFNOP", unix.NOTE_FFNOP},
{"NOTE_FFOR", unix.NOTE_FFOR},
{"NOTE_FORK", unix.NOTE_FORK},
{"NOTE_LINK", unix.NOTE_LINK},
{"NOTE_LOWAT", unix.NOTE_LOWAT},
{"NOTE_OOB", unix.NOTE_OOB},
{"NOTE_PCTRLMASK", unix.NOTE_PCTRLMASK},
{"NOTE_PDATAMASK", unix.NOTE_PDATAMASK},
{"NOTE_RENAME", unix.NOTE_RENAME},
{"NOTE_REVOKE", unix.NOTE_REVOKE},
{"NOTE_TRACK", unix.NOTE_TRACK},
{"NOTE_TRACKERR", unix.NOTE_TRACKERR},
{"NOTE_TRIGGER", unix.NOTE_TRIGGER},
{"NOTE_WRITE", unix.NOTE_WRITE},
}

View file

@ -0,0 +1,42 @@
package internal
import "golang.org/x/sys/unix"
var names = []struct {
n string
m uint32
}{
{"NOTE_ABSTIME", unix.NOTE_ABSTIME},
{"NOTE_ATTRIB", unix.NOTE_ATTRIB},
{"NOTE_CHILD", unix.NOTE_CHILD},
{"NOTE_CLOSE", unix.NOTE_CLOSE},
{"NOTE_CLOSE_WRITE", unix.NOTE_CLOSE_WRITE},
{"NOTE_DELETE", unix.NOTE_DELETE},
{"NOTE_EXEC", unix.NOTE_EXEC},
{"NOTE_EXIT", unix.NOTE_EXIT},
{"NOTE_EXTEND", unix.NOTE_EXTEND},
{"NOTE_FFAND", unix.NOTE_FFAND},
{"NOTE_FFCOPY", unix.NOTE_FFCOPY},
{"NOTE_FFCTRLMASK", unix.NOTE_FFCTRLMASK},
{"NOTE_FFLAGSMASK", unix.NOTE_FFLAGSMASK},
{"NOTE_FFNOP", unix.NOTE_FFNOP},
{"NOTE_FFOR", unix.NOTE_FFOR},
{"NOTE_FILE_POLL", unix.NOTE_FILE_POLL},
{"NOTE_FORK", unix.NOTE_FORK},
{"NOTE_LINK", unix.NOTE_LINK},
{"NOTE_LOWAT", unix.NOTE_LOWAT},
{"NOTE_MSECONDS", unix.NOTE_MSECONDS},
{"NOTE_NSECONDS", unix.NOTE_NSECONDS},
{"NOTE_OPEN", unix.NOTE_OPEN},
{"NOTE_PCTRLMASK", unix.NOTE_PCTRLMASK},
{"NOTE_PDATAMASK", unix.NOTE_PDATAMASK},
{"NOTE_READ", unix.NOTE_READ},
{"NOTE_RENAME", unix.NOTE_RENAME},
{"NOTE_REVOKE", unix.NOTE_REVOKE},
{"NOTE_SECONDS", unix.NOTE_SECONDS},
{"NOTE_TRACK", unix.NOTE_TRACK},
{"NOTE_TRACKERR", unix.NOTE_TRACKERR},
{"NOTE_TRIGGER", unix.NOTE_TRIGGER},
{"NOTE_USECONDS", unix.NOTE_USECONDS},
{"NOTE_WRITE", unix.NOTE_WRITE},
}

View file

@ -0,0 +1,32 @@
//go:build freebsd || openbsd || netbsd || dragonfly || darwin
package internal
import (
"fmt"
"os"
"strings"
"time"
"golang.org/x/sys/unix"
)
func Debug(name string, kevent *unix.Kevent_t) {
mask := uint32(kevent.Fflags)
var (
l []string
unknown = mask
)
for _, n := range names {
if mask&n.m == n.m {
l = append(l, n.n)
unknown ^= n.m
}
}
if unknown > 0 {
l = append(l, fmt.Sprintf("0x%x", unknown))
}
fmt.Fprintf(os.Stderr, "FSNOTIFY_DEBUG: %s %10d:%-60s → %q\n",
time.Now().Format("15:04:05.000000000"), mask, strings.Join(l, " | "), name)
}

View file

@ -0,0 +1,56 @@
package internal
import (
"fmt"
"os"
"strings"
"time"
"golang.org/x/sys/unix"
)
func Debug(name string, mask, cookie uint32) {
names := []struct {
n string
m uint32
}{
{"IN_ACCESS", unix.IN_ACCESS},
{"IN_ATTRIB", unix.IN_ATTRIB},
{"IN_CLOSE", unix.IN_CLOSE},
{"IN_CLOSE_NOWRITE", unix.IN_CLOSE_NOWRITE},
{"IN_CLOSE_WRITE", unix.IN_CLOSE_WRITE},
{"IN_CREATE", unix.IN_CREATE},
{"IN_DELETE", unix.IN_DELETE},
{"IN_DELETE_SELF", unix.IN_DELETE_SELF},
{"IN_IGNORED", unix.IN_IGNORED},
{"IN_ISDIR", unix.IN_ISDIR},
{"IN_MODIFY", unix.IN_MODIFY},
{"IN_MOVE", unix.IN_MOVE},
{"IN_MOVED_FROM", unix.IN_MOVED_FROM},
{"IN_MOVED_TO", unix.IN_MOVED_TO},
{"IN_MOVE_SELF", unix.IN_MOVE_SELF},
{"IN_OPEN", unix.IN_OPEN},
{"IN_Q_OVERFLOW", unix.IN_Q_OVERFLOW},
{"IN_UNMOUNT", unix.IN_UNMOUNT},
}
var (
l []string
unknown = mask
)
for _, n := range names {
if mask&n.m == n.m {
l = append(l, n.n)
unknown ^= n.m
}
}
if unknown > 0 {
l = append(l, fmt.Sprintf("0x%x", unknown))
}
var c string
if cookie > 0 {
c = fmt.Sprintf("(cookie: %d) ", cookie)
}
fmt.Fprintf(os.Stderr, "FSNOTIFY_DEBUG: %s %-30s → %s%q\n",
time.Now().Format("15:04:05.000000000"), strings.Join(l, "|"), c, name)
}

View file

@ -0,0 +1,25 @@
package internal
import "golang.org/x/sys/unix"
var names = []struct {
n string
m uint32
}{
{"NOTE_ATTRIB", unix.NOTE_ATTRIB},
{"NOTE_CHILD", unix.NOTE_CHILD},
{"NOTE_DELETE", unix.NOTE_DELETE},
{"NOTE_EXEC", unix.NOTE_EXEC},
{"NOTE_EXIT", unix.NOTE_EXIT},
{"NOTE_EXTEND", unix.NOTE_EXTEND},
{"NOTE_FORK", unix.NOTE_FORK},
{"NOTE_LINK", unix.NOTE_LINK},
{"NOTE_LOWAT", unix.NOTE_LOWAT},
{"NOTE_PCTRLMASK", unix.NOTE_PCTRLMASK},
{"NOTE_PDATAMASK", unix.NOTE_PDATAMASK},
{"NOTE_RENAME", unix.NOTE_RENAME},
{"NOTE_REVOKE", unix.NOTE_REVOKE},
{"NOTE_TRACK", unix.NOTE_TRACK},
{"NOTE_TRACKERR", unix.NOTE_TRACKERR},
{"NOTE_WRITE", unix.NOTE_WRITE},
}

View file

@ -0,0 +1,28 @@
package internal
import "golang.org/x/sys/unix"
var names = []struct {
n string
m uint32
}{
{"NOTE_ATTRIB", unix.NOTE_ATTRIB},
// {"NOTE_CHANGE", unix.NOTE_CHANGE}, // Not on 386?
{"NOTE_CHILD", unix.NOTE_CHILD},
{"NOTE_DELETE", unix.NOTE_DELETE},
{"NOTE_EOF", unix.NOTE_EOF},
{"NOTE_EXEC", unix.NOTE_EXEC},
{"NOTE_EXIT", unix.NOTE_EXIT},
{"NOTE_EXTEND", unix.NOTE_EXTEND},
{"NOTE_FORK", unix.NOTE_FORK},
{"NOTE_LINK", unix.NOTE_LINK},
{"NOTE_LOWAT", unix.NOTE_LOWAT},
{"NOTE_PCTRLMASK", unix.NOTE_PCTRLMASK},
{"NOTE_PDATAMASK", unix.NOTE_PDATAMASK},
{"NOTE_RENAME", unix.NOTE_RENAME},
{"NOTE_REVOKE", unix.NOTE_REVOKE},
{"NOTE_TRACK", unix.NOTE_TRACK},
{"NOTE_TRACKERR", unix.NOTE_TRACKERR},
{"NOTE_TRUNCATE", unix.NOTE_TRUNCATE},
{"NOTE_WRITE", unix.NOTE_WRITE},
}

View file

@ -0,0 +1,45 @@
package internal
import (
"fmt"
"os"
"strings"
"time"
"golang.org/x/sys/unix"
)
func Debug(name string, mask int32) {
names := []struct {
n string
m int32
}{
{"FILE_ACCESS", unix.FILE_ACCESS},
{"FILE_MODIFIED", unix.FILE_MODIFIED},
{"FILE_ATTRIB", unix.FILE_ATTRIB},
{"FILE_TRUNC", unix.FILE_TRUNC},
{"FILE_NOFOLLOW", unix.FILE_NOFOLLOW},
{"FILE_DELETE", unix.FILE_DELETE},
{"FILE_RENAME_TO", unix.FILE_RENAME_TO},
{"FILE_RENAME_FROM", unix.FILE_RENAME_FROM},
{"UNMOUNTED", unix.UNMOUNTED},
{"MOUNTEDOVER", unix.MOUNTEDOVER},
{"FILE_EXCEPTION", unix.FILE_EXCEPTION},
}
var (
l []string
unknown = mask
)
for _, n := range names {
if mask&n.m == n.m {
l = append(l, n.n)
unknown ^= n.m
}
}
if unknown > 0 {
l = append(l, fmt.Sprintf("0x%x", unknown))
}
fmt.Fprintf(os.Stderr, "FSNOTIFY_DEBUG: %s %10d:%-30s → %q\n",
time.Now().Format("15:04:05.000000000"), mask, strings.Join(l, " | "), name)
}

View file

@ -0,0 +1,40 @@
package internal
import (
"fmt"
"os"
"path/filepath"
"strings"
"time"
"golang.org/x/sys/windows"
)
func Debug(name string, mask uint32) {
names := []struct {
n string
m uint32
}{
{"FILE_ACTION_ADDED", windows.FILE_ACTION_ADDED},
{"FILE_ACTION_REMOVED", windows.FILE_ACTION_REMOVED},
{"FILE_ACTION_MODIFIED", windows.FILE_ACTION_MODIFIED},
{"FILE_ACTION_RENAMED_OLD_NAME", windows.FILE_ACTION_RENAMED_OLD_NAME},
{"FILE_ACTION_RENAMED_NEW_NAME", windows.FILE_ACTION_RENAMED_NEW_NAME},
}
var (
l []string
unknown = mask
)
for _, n := range names {
if mask&n.m == n.m {
l = append(l, n.n)
unknown ^= n.m
}
}
if unknown > 0 {
l = append(l, fmt.Sprintf("0x%x", unknown))
}
fmt.Fprintf(os.Stderr, "FSNOTIFY_DEBUG: %s %-65s → %q\n",
time.Now().Format("15:04:05.000000000"), strings.Join(l, " | "), filepath.ToSlash(name))
}

View file

@ -0,0 +1,31 @@
//go:build freebsd
package internal
import (
"syscall"
"golang.org/x/sys/unix"
)
var (
SyscallEACCES = syscall.EACCES
UnixEACCES = unix.EACCES
)
var maxfiles uint64
func SetRlimit() {
// Go 1.19 will do this automatically: https://go-review.googlesource.com/c/go/+/393354/
var l syscall.Rlimit
err := syscall.Getrlimit(syscall.RLIMIT_NOFILE, &l)
if err == nil && l.Cur != l.Max {
l.Cur = l.Max
syscall.Setrlimit(syscall.RLIMIT_NOFILE, &l)
}
maxfiles = uint64(l.Cur)
}
func Maxfiles() uint64 { return maxfiles }
func Mkfifo(path string, mode uint32) error { return unix.Mkfifo(path, mode) }
func Mknod(path string, mode uint32, dev int) error { return unix.Mknod(path, mode, uint64(dev)) }

View file

@ -0,0 +1,2 @@
// Package internal contains some helpers.
package internal

31
vendor/github.com/fsnotify/fsnotify/internal/unix.go generated vendored Normal file
View file

@ -0,0 +1,31 @@
//go:build !windows && !darwin && !freebsd
package internal
import (
"syscall"
"golang.org/x/sys/unix"
)
var (
SyscallEACCES = syscall.EACCES
UnixEACCES = unix.EACCES
)
var maxfiles uint64
func SetRlimit() {
// Go 1.19 will do this automatically: https://go-review.googlesource.com/c/go/+/393354/
var l syscall.Rlimit
err := syscall.Getrlimit(syscall.RLIMIT_NOFILE, &l)
if err == nil && l.Cur != l.Max {
l.Cur = l.Max
syscall.Setrlimit(syscall.RLIMIT_NOFILE, &l)
}
maxfiles = uint64(l.Cur)
}
func Maxfiles() uint64 { return maxfiles }
func Mkfifo(path string, mode uint32) error { return unix.Mkfifo(path, mode) }
func Mknod(path string, mode uint32, dev int) error { return unix.Mknod(path, mode, dev) }

View file

@ -0,0 +1,7 @@
//go:build !windows
package internal
func HasPrivilegesForSymlink() bool {
return true
}

View file

@ -0,0 +1,41 @@
//go:build windows
package internal
import (
"errors"
"golang.org/x/sys/windows"
)
// Just a dummy.
var (
SyscallEACCES = errors.New("dummy")
UnixEACCES = errors.New("dummy")
)
func SetRlimit() {}
func Maxfiles() uint64 { return 1<<64 - 1 }
func Mkfifo(path string, mode uint32) error { return errors.New("no FIFOs on Windows") }
func Mknod(path string, mode uint32, dev int) error { return errors.New("no device nodes on Windows") }
func HasPrivilegesForSymlink() bool {
var sid *windows.SID
err := windows.AllocateAndInitializeSid(
&windows.SECURITY_NT_AUTHORITY,
2,
windows.SECURITY_BUILTIN_DOMAIN_RID,
windows.DOMAIN_ALIAS_RID_ADMINS,
0, 0, 0, 0, 0, 0,
&sid)
if err != nil {
return false
}
defer windows.FreeSid(sid)
token := windows.Token(0)
member, err := token.IsMember(sid)
if err != nil {
return false
}
return member || token.IsElevated()
}

View file

@ -1,208 +0,0 @@
#!/usr/bin/env zsh
[ "${ZSH_VERSION:-}" = "" ] && echo >&2 "Only works with zsh" && exit 1
setopt err_exit no_unset pipefail extended_glob
# Simple script to update the godoc comments on all watchers. Probably took me
# more time to write this than doing it manually, but ah well 🙃
watcher=$(<<EOF
// Watcher watches a set of paths, delivering events on a channel.
//
// A watcher should not be copied (e.g. pass it by pointer, rather than by
// value).
//
// # Linux notes
//
// When a file is removed a Remove event won't be emitted until all file
// descriptors are closed, and deletes will always emit a Chmod. For example:
//
// fp := os.Open("file")
// os.Remove("file") // Triggers Chmod
// fp.Close() // Triggers Remove
//
// This is the event that inotify sends, so not much can be changed about this.
//
// The fs.inotify.max_user_watches sysctl variable specifies the upper limit
// for the number of watches per user, and fs.inotify.max_user_instances
// specifies the maximum number of inotify instances per user. Every Watcher you
// create is an "instance", and every path you add is a "watch".
//
// These are also exposed in /proc as /proc/sys/fs/inotify/max_user_watches and
// /proc/sys/fs/inotify/max_user_instances
//
// To increase them you can use sysctl or write the value to the /proc file:
//
// # Default values on Linux 5.18
// sysctl fs.inotify.max_user_watches=124983
// sysctl fs.inotify.max_user_instances=128
//
// To make the changes persist on reboot edit /etc/sysctl.conf or
// /usr/lib/sysctl.d/50-default.conf (details differ per Linux distro; check
// your distro's documentation):
//
// fs.inotify.max_user_watches=124983
// fs.inotify.max_user_instances=128
//
// Reaching the limit will result in a "no space left on device" or "too many open
// files" error.
//
// # kqueue notes (macOS, BSD)
//
// kqueue requires opening a file descriptor for every file that's being watched;
// so if you're watching a directory with five files then that's six file
// descriptors. You will run in to your system's "max open files" limit faster on
// these platforms.
//
// The sysctl variables kern.maxfiles and kern.maxfilesperproc can be used to
// control the maximum number of open files, as well as /etc/login.conf on BSD
// systems.
//
// # macOS notes
//
// Spotlight indexing on macOS can result in multiple events (see [#15]). A
// temporary workaround is to add your folder(s) to the "Spotlight Privacy
// Settings" until we have a native FSEvents implementation (see [#11]).
//
// [#11]: https://github.com/fsnotify/fsnotify/issues/11
// [#15]: https://github.com/fsnotify/fsnotify/issues/15
EOF
)
new=$(<<EOF
// NewWatcher creates a new Watcher.
EOF
)
add=$(<<EOF
// Add starts monitoring the path for changes.
//
// A path can only be watched once; attempting to watch it more than once will
// return an error. Paths that do not yet exist on the filesystem cannot be
// added. A watch will be automatically removed if the path is deleted.
//
// A path will remain watched if it gets renamed to somewhere else on the same
// filesystem, but the monitor will get removed if the path gets deleted and
// re-created, or if it's moved to a different filesystem.
//
// Notifications on network filesystems (NFS, SMB, FUSE, etc.) or special
// filesystems (/proc, /sys, etc.) generally don't work.
//
// # Watching directories
//
// All files in a directory are monitored, including new files that are created
// after the watcher is started. Subdirectories are not watched (i.e. it's
// non-recursive).
//
// # Watching files
//
// Watching individual files (rather than directories) is generally not
// recommended as many tools update files atomically. Instead of "just" writing
// to the file a temporary file will be written to first, and if successful the
// temporary file is moved to to destination removing the original, or some
// variant thereof. The watcher on the original file is now lost, as it no
// longer exists.
//
// Instead, watch the parent directory and use Event.Name to filter out files
// you're not interested in. There is an example of this in [cmd/fsnotify/file.go].
EOF
)
remove=$(<<EOF
// Remove stops monitoring the path for changes.
//
// Directories are always removed non-recursively. For example, if you added
// /tmp/dir and /tmp/dir/subdir then you will need to remove both.
//
// Removing a path that has not yet been added returns [ErrNonExistentWatch].
EOF
)
close=$(<<EOF
// Close removes all watches and closes the events channel.
EOF
)
watchlist=$(<<EOF
// WatchList returns all paths added with [Add] (and are not yet removed).
EOF
)
events=$(<<EOF
// Events sends the filesystem change events.
//
// fsnotify can send the following events; a "path" here can refer to a
// file, directory, symbolic link, or special file like a FIFO.
//
// fsnotify.Create A new path was created; this may be followed by one
// or more Write events if data also gets written to a
// file.
//
// fsnotify.Remove A path was removed.
//
// fsnotify.Rename A path was renamed. A rename is always sent with the
// old path as Event.Name, and a Create event will be
// sent with the new name. Renames are only sent for
// paths that are currently watched; e.g. moving an
// unmonitored file into a monitored directory will
// show up as just a Create. Similarly, renaming a file
// to outside a monitored directory will show up as
// only a Rename.
//
// fsnotify.Write A file or named pipe was written to. A Truncate will
// also trigger a Write. A single "write action"
// initiated by the user may show up as one or multiple
// writes, depending on when the system syncs things to
// disk. For example when compiling a large Go program
// you may get hundreds of Write events, so you
// probably want to wait until you've stopped receiving
// them (see the dedup example in cmd/fsnotify).
//
// fsnotify.Chmod Attributes were changed. On Linux this is also sent
// when a file is removed (or more accurately, when a
// link to an inode is removed). On kqueue it's sent
// and on kqueue when a file is truncated. On Windows
// it's never sent.
EOF
)
errors=$(<<EOF
// Errors sends any errors.
EOF
)
set-cmt() {
local pat=$1
local cmt=$2
IFS=$'\n' local files=($(grep -n $pat backend_*~*_test.go))
for f in $files; do
IFS=':' local fields=($=f)
local file=$fields[1]
local end=$(( $fields[2] - 1 ))
# Find start of comment.
local start=0
IFS=$'\n' local lines=($(head -n$end $file))
for (( i = 1; i <= $#lines; i++ )); do
local line=$lines[-$i]
if ! grep -q '^[[:space:]]*//' <<<$line; then
start=$(( end - (i - 2) ))
break
fi
done
head -n $(( start - 1 )) $file >/tmp/x
print -r -- $cmt >>/tmp/x
tail -n+$(( end + 1 )) $file >>/tmp/x
mv /tmp/x $file
done
}
set-cmt '^type Watcher struct ' $watcher
set-cmt '^func NewWatcher(' $new
set-cmt '^func (w \*Watcher) Add(' $add
set-cmt '^func (w \*Watcher) Remove(' $remove
set-cmt '^func (w \*Watcher) Close(' $close
set-cmt '^func (w \*Watcher) WatchList(' $watchlist
set-cmt '^[[:space:]]*Events *chan Event$' $events
set-cmt '^[[:space:]]*Errors *chan error$' $errors

View file

@ -1,5 +1,4 @@
//go:build freebsd || openbsd || netbsd || dragonfly
// +build freebsd openbsd netbsd dragonfly
package fsnotify

View file

@ -1,5 +1,4 @@
//go:build darwin
// +build darwin
package fsnotify